Understanding the risks of social exclusion across …...The analysis and views expressed in this...

The analysis and views expressed in this paper are not a statement of Government policy.

Understanding the risks of social exclusion across the life course: Older age Elizabeth Becker and Richard Boreham National Centre for Social Research

The analysis and views expressed in this paper are not a statement of Government policy.

Acknowledgements The data for this study were made available through the UK Data Archive (UKDA). The British Household Panel Survey (BHPS) is conducted by the Economic and Social Research Council (ESRC) UK Longitudinal Studies Centre (ULSC), together with the Institute for Social and Economic Research (ISER) at the University of Essex. The data were collected by GfK NOP, Office of National Statistics and Northern Ireland Statistics and Research Agency. The funding is provided by the Economic and Social Research Council. ELSA was developed by a team of researchers based at the National Centre for Social Research, University College London and the Institute for Fiscal Studies. The data were collected by the National Centre for Social Research. The funding is provided by the National Institute of Aging in the United States, and a consortium of UK government departments co-ordinated by the Office for National Statistics. The developers and funders of BHPS, ELSA and the Archive do not bear any responsibility for the analyses or interpretations presented here. The authors would like to thank colleagues at NatCen for their contribution to this report. For her help in preparing the ELSA data for analysis we would like to thank Jenny Godlieb. For her help with checking the final version of this report we would like to thank Jenny Chanfreau. We would also like to thank John D’Souza for his advice on the analysis of both the ELSA and BHPS data.

CONTENTS List of figures .....................................................................................................................................i List of tables ....................................................................................................................................ii EXECUTIVE SUMMARY .................................................................................................................. 1 1 Introduction ................................................................................................................................ 4 2 Multiple and multidimensional risk markers of social exclusion ............................................... 14

2.1 The multidimensional nature social exclusion ................................................................. 14 2.2 Which risk markers do older people experience? ........................................................... 16

2.2.1 Latent Class Analysis findings........................................................................ 17 Cluster One: Mostly lonely and unsupported.................................................. 17 Cluster Two: Fear of local area ...................................................................... 19 Cluster Three: Low income............................................................................. 20 Cluster Four: Mostly poor health .................................................................... 22 Cluster Five: Mostly poor access.................................................................... 23

2.2.2 Comparing clusters of multiple risk markers of social exclusion .................... 24 2.3 How do clusters of multiple risk markers of social exclusion vary?................................. 27

Demographic characteristics .......................................................................... 27 Home and area characteristics....................................................................... 32 Health factors ................................................................................................. 35 Summary ........................................................................................................ 37

3 Social exclusion risk marker dynamics .................................................................................... 38

3.1 Measuring the dynamics of social exclusion ................................................................... 38 3.2 Profile of the balanced panel sample .............................................................................. 40 3.3 The behaviour of multidimensional social exclusion over time........................................ 42

Low social support and loneliness cluster ...................................................... 44 Low income and transport cluster................................................................... 47 Health and loneliness cluster.......................................................................... 50

3.4 Comparing combinations of multiple risk markers of social exclusion over time............. 52 3.5 Triggers of becoming marked by risk .............................................................................. 55

4 Policy directions ....................................................................................................................... 57 REFERENCES ............................................................................................................................... 58 5 Annex A: Tables....................................................................................................................... 60

5.1 Markers of risk by age and sex ....................................................................................... 60 5.2 Socio-demographic characteristics of ELSA sample....................................................... 65 5.3 Exploring multidimensional social exclusion ................................................................... 69 5.4 Profile of the balanced panel sample .............................................................................. 74 5.5 Panel sample marker of risk 1997-2005.......................................................................... 77 5.6 Risk marker prevalence within clusters ........................................................................... 78 5.7 Socio-demographic characteristics of dynamic social exclusion clusters ....................... 80 5.8 Duration of multidimensional social exclusion................................................................. 82

5.9 Persistence in clusters: cluster transitions ...................................................................... 83 6 Annex B: Technical .................................................................................................................. 84

6.1 Correlation analysis......................................................................................................... 84 6.2 Detailing the Latent Class Analysis ................................................................................. 86

6.2.1 Latent Class Analysis ..................................................................................... 86 6.2.2 Latent GOLD .................................................................................................. 86 6.2.3 Modelling multiple markers of risk .................................................................. 87

Features of the data ....................................................................................... 87 Identifying the number of classes ................................................................... 87 Classifying individuals and describing classes ............................................... 90

6.2.4 Using Latent Class Analysis longitudinally ..................................................... 92

List of figures Figure 1-1 Sex and age profile of the sample ............................................................................. 8 Figure 1-2 Prevalence of risk markers of social exclusion ........................................................ 11 Figure 1-3 Proportion of those experiencing each risk marker who were women .................... 12 Figure 1-4 Proportion of those experiencing each risk marker who were aged 80+ ................. 13 Figure 2-1 Sex and age profile of those experiencing multiple risk markers............................. 15 Figure 2-2 Cluster One: Lonely - probabilities of risk markers ................................................. 18 Figure 2-3 Cluster Two: Fear of local area - probabilities of risk markers................................ 20 Figure 2-4 Cluster Three: Low income - probabilities of risk markers ...................................... 21 Figure 2-5 Cluster Four: Mostly poor health - probabilities of risk markers.............................. 22 Figure 2-6 Cluster Five: Mostly poor access - probabilities of risk markers ............................. 23 Figure 2-7 Multidimensional social exclusion clusters by sex ................................................... 27 Figure 2-8 Multidimensional social exclusion by age ................................................................ 28 Figure 2-9 Multidimensional social exclusion by martial status................................................. 29 Figure 2-10 Multidimensional social exclusion by living arrangements (not married) ................. 30 Figure 2-11 Multidimensional social exclusion by whether has qualifications............................. 31 Figure 2-12 Multidimensional social exclusion by whether owns own home .............................. 32 Figure 2-13 Multiple markers of risk by locality ........................................................................... 33 Figure 2-14 Multidimensional social exclusion by Index of Multiple Deprivation Quintile............ 34 Figure 2-15 Multidimensional social exclusion by whether have limiting longstanding illness.... 35 Figure 2-16 Multidimensional social exclusion by wellbeing ....................................................... 36 Figure 2-17 Multidimensional social exclusion by caring ............................................................ 37 Figure 3-1 Relationship between ELSA and BHPS multidimensional clusters ......................... 42 Figure 3-2 Cluster One: Low social support and loneliness cluster - probabilities of risk

markers.................................................................................................................... 44 Figure 3-3 Duration in low social support and loneliness cluster .............................................. 45 Figure 3-4 Persistence in low social support and loneliness cluster ......................................... 46 Figure 3-5 Cluster Two: Low income and transport cluster - probabilities of risk markers....... 47 Figure 3-6 Duration in low income and transport cluster........................................................... 48 Figure 3-7 Persistence in low income and transport cluster ..................................................... 49 Figure 3-8 Cluster Three: Health and loneliness cluster - probabilities of risk markers ........... 50 Figure 3-9 Duration in health and loneliness cluster ................................................................. 51 Figure 3-10 Persistence in health and loneliness cluster............................................................ 52 Figure 3-11 Movement to different clusters at t+1....................................................................... 53 Figure 3-12 Persistence of individual risk markers at t+1 ........................................................... 54

i Understanding the risks of social exclusion in older age List of Figures

List of tables

Table 1-1 Bristol Social Exclusion Matrix (B-SEM): risk markers of social exclusion covered in ELSA 5 Table 1-2 Risk markers.............................................................................................................. 9 Table 2-1 Risk markers used in multidimensional social exclusion analysis ........................... 14 Table 2-2 Prevalence of each risk marker of social exclusion by cluster................................. 25 Table 3-1 B-SEM risk markers of social exclusion................................................................... 38 Table 3-2 Risk markers used in longitudinal multidimensional analysis .................................. 40 Table 3-3 Prevalence of risk markers experienced by panel sample in 1997.......................... 41 Table 3-4 Duration in low social support & loneliness cluster.................................................. 45 Table 3-5 Duration in low income and transport cluster........................................................... 48 Table 3-6 Duration in health and loneliness cluster ................................................................. 51 Table 3-7 Becoming marked by risk by trigger event............................................................... 55 Table 5-1 Low income, by age and sex ................................................................................... 60 Table 5-2 Material deprivation, by age and sex ....................................................................... 60 Table 5-3 Pension wealth, by age and sex .............................................................................. 60 Table 5-4 Access to services, by age and sex......................................................................... 61 Table 5-5 Transport access, by age and sex........................................................................... 61 Table 5-6 Financial services, by age and sex.......................................................................... 61 Table 5-7 Social support, by age and sex................................................................................ 62 Table 5-8 Contact with other people, by age and sex.............................................................. 62 Table 5-9 Literacy and numeracy skills, by age and sex ......................................................... 62 Table 5-10 Political efficacy (not voted), by age and sex........................................................... 63 Table 5-11 Poor general health, by age and sex ....................................................................... 63 Table 5-12 Poor emotional health, by age and sex ................................................................... 63 Table 5-13 Low physical activity, by age and sex...................................................................... 64 Table 5-14 Bad housing, by age and sex .................................................................................. 64 Table 5-15 Sense of belonging in local area, by age and sex ................................................... 64 Table 5-16 Fear of local area after dark, by age and sex .......................................................... 65 Table 5-17 Marital status, by age and sex ................................................................................. 65 Table 5-18 Lives alone, by age and sex .................................................................................... 65 Table 5-19 Number of children, by age and sex ........................................................................ 66 Table 5-20 Number of siblings, by age and sex......................................................................... 66 Table 5-21 Qualifications, by age and sex................................................................................. 66 Table 5-22 Housing tenure, by age and sex .............................................................................. 67 Table 5-23 Limiting longstanding illness, by age and sex.......................................................... 67 Table 5-24 Housing tenure, by age and sex .............................................................................. 67 Table 5-25 Index of Multiple Deprivation Quintile, by age and sex............................................ 68 Table 5-26 Well-being, by age and sex ..................................................................................... 68 Table 5-27 Carer, by age and sex ............................................................................................. 68 Table 5-28 Population density, by age and sex ......................................................................... 69 Table 5-29 Ethnicity, by age and sex......................................................................................... 69 Table 5-30 Multidimensional social exclusion, by sex ............................................................... 69 Table 5-31 Multidimensional social exclusion, by age ............................................................... 70 Table 5-32 Multidimensional social exclusion, by marital status................................................ 70 Table 5-33 Multidimensional social exclusion, by whether lives alone ...................................... 70 Table 5-34 Multidimensional social exclusion, by whether has living siblings ........................... 71

ii Understanding the risks of social exclusion in older age List of Tables

Table 5-35 Multidimensional social exclusion, by whether has qualifications............................ 71 Table 5-36 Multidimensional social exclusion, by tenure........................................................... 72 Table 5-37 Multidimensional social exclusion, by income quintile ............................................. 72 Table 5-38 Multidimensional social exclusion, by whether has limiting longstanding illness..... 72 Table 5-39 Multidimensional social exclusion, by IMD quintile .................................................. 73 Table 5-40 Multidimensional social exclusion, by well-being..................................................... 73 Table 5-41 Multidimensional social exclusion, by whether is a carer ........................................ 73 Table 5-42 Multidimensional social exclusion, by population density ........................................ 74 Table 5-43 Age and sex profile of the balanced panel sample .................................................. 74 Table 5-44 Marital status, by sex and wave............................................................................... 75 Table 5-45 Employment in the household, by sex and wave..................................................... 75 Table 5-46 Population density, by sex at wave 1997................................................................. 76 Table 5-47 Index of Multiple Deprivation, by sex at wave 1997................................................. 76 Table 5-48 Balanced panel sample: prevalence of risk markers ............................................... 77 Table 5-49 Dynamic social exclusion clusters, by year ............................................................ 78 Table 5-49 continued Dynamic social exclusion clusters, by year ............................................ 79 Table 5-50 Dynamic social exclusion clusters, by sex............................................................. 80 Table 5-51 Dynamic social exclusion clusters, by age .............................................................. 80 Table 5.52 Dynamic social exclusion clusters, by family type ................................................... 80 Table 5-53 Dynamic social exclusion clusters, by marital status ............................................... 80 Table 5-54 Dynamic social exclusion clusters, by housing tenure............................................. 81 Table 5-55 Dynamic social exclusion clusters, by population density at first wave ................... 81 Table 5.56 Dynamic social exclusion clusters, by Index of Multiple Deprivation at first wave ... 81 Table 5-57 Duration of low social support and loneliness multiple risk markers........................ 82 Table 5-58 Duration of low income and transport multiple risk markers .................................... 82 Table 5-59 Duration of health and loneliness multiple risk markers .......................................... 82 Table 5-60 Persistence and movement of Low social support and loneliness ......................... 83 Table 5-61 Persistence and movement of Low income and transport Problems....................... 83 Table 5-62 Persistence and movement of Health and loneliness............................................. 83 Table 6-1 Tetrachoric Correlation Matrix ....................................................................................... 85 Table 6-2 Latent Class Analysis models and goodness-of-fit statistics ......................................... 89 Table 6-3 Conditional probability parameters of Latent Class Analysis model .............................. 91 Table 6-4 Latent Class Analysis model probabilities across 4 waves of BHPS............................. 92

Understanding the risks of social exclusion in older age Executive summary 1

EXECUTIVE SUMMARY Social exclusion means being unable to access the things in life that most of society takes for granted. It’s not just about having enough money. It is a build-up of problems across several aspects of people’s lives (Age Concern 2009). Older people can face a range of problems which can be thought of as risk markers of social exclusion, for example low income, limited contact with other people, and poor health.

Previous studies have focused on singular risk markers of social exclusion. This study focused older people aged 60 and over, who experienced multiple risk markers, as evidence suggests that experiencing two or more risk markers of social exclusion can have severe negative implications and consequences for quality of life (Barnes et al 2006). The need to understand the experience of multiple risk markers in older age is very important, especially as Britain has a rapidly growing older population, with more pensioners than children recorded living in the UK in 2008.

This study was funded by the Social Exclusion Task Force, and was carried out by Elizabeth Becker and Richard Boreham of the National Centre for Social Research (NatCen). The main report presents the findings of this study. Key results are presented in this summary.

How is social exclusion measured? The study uses 2004-05 data from the English Longitudinal Study of Ageing (ELSA), a survey broad in topic area, which can be used to measure patterns of multiple risk markers of social exclusion in depth. Nine years of the British Household Panel Survey (BHPS) were used in exploring the dynamics of social exclusion risk markers. The BHPS is less detailed when considering the measurement of social exclusion risk markers, but is rich in its longitudinal nature. In both data sources, social exclusion was measured using the risk markers outlined in the Bristol Social Exclusion Matrix (Levitas et al. 2007).

Who experiences risk markers of social exclusion? Findings from this study showed that 50 per cent of all those aged 60 and older experienced multiple risk markers of social exclusion.1 The older old (aged 80 years and over) were more likely to experience multiple risk markers than their younger counterparts (72% of those aged 80 and over, compared with 52% of those aged 70-79, and 41% of those aged 60-69 experienced multiple risk markers).

What are the combinations of multiple risk markers of social exclusion?

Older people who experienced multiple risk markers were not a homogenous group in terms of the problems they faced. Latent Class Analysis, a technique which sought to understand how risk

1 Experiencing multiple risk markers of social exclusion was defined as having two or more of 16 measures created in the ELSA data. These 16 risk markers were based on those presented in the Bristol Social Exclusion Matrix (Levitas et al 2007).


markers grouped together, was used to identify different combinations of risk markers experienced by older people. 2

Findings from this research suggest that:

• On average, 5% of older people experienced 5.1 risk markers of social exclusion out of 16, more than any other group.

Older people in this group were very likely to have poor access to services and to transport, were physically inactive, had a fear of their local area after dark, had low social support, and had poor general and emotional health.

Older people in this group were likely to be aged 80 and older, had no qualifications, and lived alone.

• 7% of older people experienced 3.7 risk markers of social exclusion on average. Older people in this group were likely to experience poor general and emotional health, low

social support, and had a fear of their local area after dark. Older people in this group had no qualifications, and had a longstanding and limiting

illness.

• 16% of older people experienced 3.2 risk markers of social exclusion on average. Older people in this group experienced low relative income, had a fear of their local area

after dark, and had low social support. Older people in this group were likely to be female, aged 80 and older, lived alone, were

unmarried, lived in a deprived area, and rented their home.

• 12% of older people experienced 2.7 risk markers of social exclusion on average. Older people in this group had a fear of their local area after dark, and had poor literacy and

numeracy skills. Older people in this group were likely to live in a deprived area, and rented their home.

• 10% of older people experienced 2.5 risk markers of social exclusion on average. Older people in this group had limited contact with others, and had low social support. Older people in this group were likely to be unmarried.

How does social exclusion behave over time? Nine years of BHPS data was used to explore the dynamics of multiple risk markers of social exclusion. Three combinations, in other words clusters, of multiple risk markers were studied over time3: • a combination of low income and transport access risk markers, • a combination of poor health and loneliness risk markers, • a combination of low social support and loneliness risk markers.

2 Latent Class Analysis was used to identify clusters of multiple risk markers. Technical details of this analysis are presented in Annex B. A caveat to the Latent Class Analysis that should be borne in mind, is that it is driven by the 16 risk markers available in the ELSA data, and this to some extent determines the combinations of risk markers identified. 3 These groups were identified through a Latent Class Analysis of BHPS data. The three clusters of multiple risk markers classified by the analysis were a somewhat compressed translation of those identified in the earlier analysis of ELSA data.


Analysis looked at how long older people experienced these combinations of multiple markers of risk. Over the nine years analysed, around two thirds (65%) of older people experienced a combination of multiple risk markers related to health and loneliness on at least one occasion, and around two thirds (64%) of older people experienced a combination of multiple risk markers associated with low income and poor transport access on at least one occasion. A larger proportion of older people (83%) experienced a combination of multiple risk markers associated with low social support and loneliness on at least one occasion. The study focussed on older people that persistently had multiple risk markers, as these people were likely to have the lowest living standards. Older people who had a combination of low income and transport risk markers experienced this set of problems the longest. 28% of older people who faced this combination of risk markers, did so for at least seven years. 10% experienced this combination of risk markers for all nine years. Older people experiencing health and loneliness problems, or social support and loneliness problems experienced these combinations of multiple risk markers for a shorter duration 4. The research also looked at whether older people moved from one combination of multiple risk markers to another. Older people who stopped experiencing a combination of health and loneliness risk markers, were likely to move out of this situation completely, rather than to a different combination of risk markers. Conversely, older people who moved from a combination of low income and transport risk markers, or a combination of low social support and loneliness risk markers, were more likely to move to a different situation of multiple risk markers, than to escape a situation of multiple risk altogether. What triggers the experience of multiple risk markers of social exclusion? In addition to exploring the behaviour of multiple risk markers over time, this study looked at the impact of life events between one year and the next, such as becoming retired. Getting divorced between one year and the next, or becoming widowed, were the only events that were significantly related to a change in situation from experiencing none or one risk marker of social exclusion to experiencing multiple risk markers. No life event triggered a change in situation from one combination of multiple risk markers to a different combination of multiple risk markers.

4 The persistence of experiencing these forms of multiple risk markers is probably due to the persistence of these markers as individual measures of disadvantage, rather than of multiple risk markers per se.

Understanding the risks of social exclusion in older age Introduction 4

1 Introduction 1.1 Conceptualising social exclusion The concept of social exclusion arose from a concern that poverty was too one-dimensional as a measurement of disadvantage, and that there were other factors that should be taken into account when creating policy to improve the situation for those who are worse off in society. Poverty measurements cover the financial or economic dimension of exclusion, and broadening the definition from poverty to social exclusion has led people to consider the social, political and cultural aspects of disadvantage. The problem with defining and measuring social exclusion is determining which factors can be considered as risk markers of social exclusion and how they can be grouped meaningfully together into dimensions or domains. There are a number of different approaches to conceptualising social exclusion (Barnes, 2005; Berthoud, 2003; Barnes et al, 2002; Burchardt et al, 1999; Lessof and Jowell, 1999 are some of the more recent studies) and it is difficult to assess which methods of conceptualising social exclusion are more valid than others. Levitas et al (2007) extended the discussion about definitions of exclusion to define an additional concept of deep exclusion, which essentially is experiencing a risk marker of social exclusion across more than one dimension or domain. The study examined the patterns of deep exclusion through an exploration of the combinations of the multiple risk markers of exclusion experienced by older people. In addition to the problem of theoretically defining social exclusion, there are problems with its measurement. The choice of which risk markers of social exclusion to use is primarily driven by the data that is already available, whether survey data or from administrative records. 1.2 Measuring social exclusion Social exclusion has been defined successfully elsewhere and a number of studies have attempted to measure and describe this important but complex concept (Levitas et al, 2007; Nolan and Whelan, 1996; Room, 1998). Helpfully, Levitas et al (2007) have created the Bristol Social Exclusion Matrix (B-SEM) which identifies ten domains of social exclusion, falling into three main themes: lack of resources, lack of participation in economic, social, cultural and political life and quality of life. This method of operationalising social exclusion, as with all other methods of this kind, makes certain assumptions about what can be considered to be the components of social exclusion as opposed to what are the risk markers and outcomes of social exclusion. For example, it could be argued that some of the subcomponents of social exclusion identified by Levitas et al, such as having a reduced income, being unemployed and having health problems (e.g. a stroke or a fall) can be considered to be drivers of social exclusion rather than components of it. Furthermore, other subcomponents, such as imprisonment, poor life satisfaction, and again health problems (e.g. asthma, depression), could be thought of as potential outcomes of being socially excluded.


The B-SEM is a hierarchical categorisation of exclusion consisting of three domains, 10 sub-domains, and 56 themes of which there are risk markers of social exclusion. The choice of risk markers in this analysis is limited by two factors, the coverage of risk markers in ELSA and the relevance of markers to older people (for example being employed is not as relevant for a population largely above retirement age). Previously studies of social exclusion have been more difficult to compare. Levitas et al (2007, page 81) said of Barnes et al (2006) “Comparing these findings to other studies of social exclusion such as the Poverty and Social Exclusion Survey (Gordon et al, 2000; Pantazis et al, 2006) and SQOL OA (Survey of Quality of Life in Old Age) (Scharf et al, 2002) is problematic because the scoring mechanisms and cut-off thresholds employed in each study are different”. All studies in this series have used the B-SEM in defining markers of risk. This means that findings in the older life stage can more easily be compared with other studies that use this framework. This analysis covers 21 risk markers of social exclusion and 9 of the B-SEM sub-domains. Table 1-1 Bristol Social Exclusion Matrix (B-SEM): risk markers of social exclusion covered in ELSA

Domain Sub-domain Risk marker of social exclusion Covered Resources Material and economic resources Relative low income Receipt of out of work benefits Material deprivation Fuel Poverty Not owning own home No pension wealth No wealth and savings Debt Subjective poverty Access to public and private services Poor access to services Poor access to transport No financial services No private services Poor utilities Social resources Institutionalisation Low social support Low contact with others Participation Economic participation Cares for another person In employment Living in a workless household Undertaking unpaid work Poor quality of working life Social Participation No participation in positive activities Culture, education and skills Poor functional literacy and numeracy School absences and exclusions Not doing cultural, leisure activities

Not participating in cultural and sporting activities

No internet access No qualifications Political and civic participation Not voted in the last general election Quality of life Health and well-being Poor self-reported general health Low participation in physical exercise


Domain Sub-domain Risk marker Covered Health and well-being Obesity Limiting longstanding illness Poor emotional health Low well being Low personal efficacy Long periods on benefits Smoking Drug use Living environment Bad housing Homelessness Poor neighbourhood safety Low area satisfaction No sense of belonging Limited access to space Crime, harm and criminalisation Fear of area after dark Experience of crime Hospital admissions Domestic violence Fear of crime Bullying Discrimination Committed a crime Imprisonment Has an ASBO

1.3 Using the English Longitudinal Study of Ageing to measure social exclusion ELSA is funded by the National Institute on Ageing and a consortium of UK government funders led by the Office for National Statistics. It is designed and carried out through collaboration between University College London, the Institute for Fiscal Studies and the National Centre for Social Research. The ELSA sample was originally drawn from households who responded to the Health Survey for England (HSE) in 1998, 1999 or 2001. Individuals were considered eligible to be core members of ELSA if they had been living in an HSE household, were 50 or over and were still living in a private residential address in England. The first ELSA survey was carried out in 2002-3. ELSA is a longitudinal study and the design strategy is to collect data every two years – wave 2 was carried out in 2004-5, and wave 3 in 2006-7. This longitudinal design means that ELSA will aid understanding of how and why people’s lives change as they grow older and in future, there will be many opportunities to look at the experiences of social exclusion over time. At the time of analysis for this study, the most recent and available wave of ELSA (wave 2) was used for cross-sectional analysis. It is important to remember that ELSA was not designed specifically to measure social exclusion, although because of its multi-disciplinary nature, a number of questions that relate to social exclusion were included in the questionnaire including individual and household characteristics; physical, cognitive, mental and psychological health; housing, work, pensions, income and assets; expectations for the future; social participation and social support.


The main drawback of using ELSA is the coverage of the population of older people. Most surveys exclude people in institutions such as prisons and people who are homeless and even those that attempt to address it struggle to achieve representative samples of these populations. This is also true of ELSA, but for the population of older people, the bigger problem is to include older people who are in care homes. ELSA’s original sample was drawn from a household sample, and although older people are followed if they move into a care home, response tends to be lower among this group, mainly due to respondents’ abilities to take part in an interview. People in care homes are clearly a group marked by risk of social exclusion, and under-represented in this analysis, although it is arguable that policies and interventions aimed at this group would be different to those aimed at the older population living in households, and that therefore the care home population should be considered separately. In summary, although there are some issues with the sample and topic coverage of ELSA, it is a rich data source and offers opportunities to explore and understand aspects of risk markers and social exclusion that would not be possible otherwise. 1.4 A profile of older people This section describes some of the basic social, demographic, and economic characteristics of older people (defined in this report as people aged 60 years and over5). It is important to recognise that older people are a diverse population, and understanding this diversity matters when deciding how to prioritise policies that might affect different groups of older people.6 In total, 55% of older people were women and 45% men. Although the population is all those aged 60 and over, the sample was biased towards the younger old – 49% were aged 60-69, 35% were aged 70-79 and 16% were aged 80 and older. Women were more likely to be older than men (18% of women were aged 80+ compared with 13% of men), and this age differential may be a factor in some of the other differences in characteristics between men and women.

5 Those aged 59 or less are explored in the associated report by Gordon, D., Fahmy,E., (2009) Understanding social exclusion across the life course: Working age, Cabinet Office. 6 A full profile of the older population can be found in Banks, J., Breeze, E., Lessof, C., Nazroo, N., (eds) (2006) Retirement, health and relationships of the older population in England: The 2004 English Longitudinal Study of Ageing. Institute of Fiscal Studies, London.


Figure 1-1 Sex and age profile of the sample

0

10

20

30

40

50

60

Men Women

Perc

ent

80+

70-79

60-69

Base: All respondents

Having significant relationships with partners or siblings are likely to be important to older people because of their effects on mortality and well-being (Barnes et al 2006). In total, 26% of older people lived alone, 22% were widowed and 26% had no living siblings, and there will obviously be some overlap between these groups. Looking at differences within the sample of older people by sex, women are more likely than men to live alone (35% compared with 15%), to be widowed (66% compared with 27%), to have a low income7 (39% compared with 26%) and for their main activity to be looking after the home (60% compared with 32%). Thus among older people, women’s lives are different to men’s lives and this may be reflected in their experience of risk markers of social exclusion. The other important factor to consider is age, particularly comparing the situation and experiences of the younger old (60-69) to the older old (80+). Comparing these two groups, the older old were more likely to be living alone (54% compared with 16%), were more likely to be widowed (58% compared with 8%) and more likely to have no living siblings (45% compared with 20%), and more likely to have a low income (36% compared with 13%). Looking at how the diverse groups within older people experience different types of markers of risk is important in terms of being able to target interventions and policies at those who would most benefit from them, and this will be explored in section 1.6.

7 Low income was defined as being in the lowest quintile of income.


1.5 Creating risk markers of social exclusion in ELSA In order to understand social exclusion and how it affects older people, we have used the framework of Levitas et al’s (2007) B-SEM, and have identified markers of risk that fit into the sub-domains and domains of the B-SEM, and that were considered relevant and appropriate to older people. The definition of risk markers of social exclusion is obviously restricted by the questions contained within ELSA. It should be noted that reported prevalence of risk markers of social exclusion are to some extent determined by the definitions of these markers as well as the number of markers used. Clearly the lower the threshold of risk, then the more people become defined as marked by risk. Similarly, the more risk markers used in the analysis, the more likely people will be defined as experiencing multiple markers of risk. Risk markers of social exclusion were defined as binary variables (marked by risk or not marked by risk) as this type of variable was required for the Latent Class Analysis, detailed later in this report. Where possible, risk markers were derived in line with government indicators. Where there were no government indicators, decisions about where to set a threshold were to some extent arbitrary. If there was a natural place to define marker of risk (such as having or not having something), then this was taken as the definition. Where there were a number of answer categories on a question, or multiple questions used to create a scale then where possible thresholds were defined such that prevalence of risk markers was in the range 15%-20%, as this was the approximate range covered by government indicators. Table 1-2 Risk markers

Material and economic resources Risk Marker Prevalence Govt Indicator Definition

Relative low income 23% PSA17 Household income is less than 60 percent of contemporary median income, before housing costs are deducted, and equivalised using the Modified OECD scale.

Material deprivation 5% OA3 Possess three or fewer of the following: central heating, freezer or fridge freezer, washing machine, microwave oven, telephone, home computer, access to internet at home.

Not owning own home 28% Living in accommodation which is shared ownership, rented, rent-free, or squatting, or buying with the help of a mortgage.

No pension wealth 5% OA22 Based on the IFS model-based estimates of pension wealth, which are broken down into state pension and private pension. A person is deemed marked by risk if their total pension wealth is zero.

Access to public and private services Risk Marker Prevalence Govt Indicator Definition

Poor access to services 13% OA7 Difficult or very difficult to access at least two of the following services: bank or cash point, chiropodist, dentist, GP, hospital, local shops, optician, post office, shopping centre, supermarket

Poor access to transport 6% OA8 No access to a car and who never or rarely use public transport.

No financial services 3% None of: Current account Savings account Tessa Isa Premium bonds National savings account, PEP, Stocks and/or shares, Share Options/Employee share ownership, Share club, Unit or investment trusts, 'Bonds and Gilts, Other savings and investments.


Social resources Risk Marker Prevalence Govt Indicator Definition

Low social support 17% Scored less than six on the derived social support scale

Low contact with others 9% OA6 Face-to-face, phone, or written contact two times a year or less with children, family, or friends.

Economic participation Risk Marker Prevalence Govt Indicator Definition

Cares for another person

13% Caring for another person, including partner or other person in or out of the household.

Culture, education and skills Risk Marker Prevalence Govt Indicator Definition

Poor functional literacy and numeracy

17% Performed poorly on both wave 2 literacy and wave 1 numeracy tests

No qualifications 45%

Political and civic participation Risk Marker Prevalence Govt Indicator Definition

Not voted in the last general election

14%

Health and well-being Risk Marker Prevalence Govt Indicator Definition

Poor self-reported general health

7% Self-reported general health is ‘Poor’

Low participation in physical exercise

6% OA16 Do mildly energetic sports and activities rarely or never.

Limiting longstanding illness

36%

Poor emotional health 11% Score at least 4 of the Center for Epidemiologic Studies Depression Scale (CES-D) symptoms

Low well being 20% Score low on the CASP-19 scale, which measures the following four elements of well-being: Control; Autonomy; Self-realisation; Pleasure

Living environment Risk Marker Prevalence Govt Indicator Definition

Bad housing 16% Have at least 1 of the following housing problems: Shortage of space; Too dark; Rising damp; Water from roof etc.; Condensation; Electrical problems; General rot and decay; Insects, mice or rats; Too cold in winter.

No sense of belonging 7% Feel to some extent that they do not belong in the area they live.

Crime, harm and criminalisation Risk Marker Prevalence Govt Indicator Definition

Fear of area after dark 29% Think to some extent that people would be afraid to walk alone in their area after dark.


Lack of qualifications was the most prevalent risk marker (45%), followed by having a longstanding limiting illness (36%), fear of area after dark (29%) and not owning a home (28%). The least prevalent risk markers were all linked to money, namely access to financial services (3%), lack of pension wealth (5%) and material deprivation (5%). Figure 1-2 Prevalence of risk markers of social exclusion

0 10 20 30 40 50

No qualif ications


Fear of crime

Not ow ning ow n home

Relative low income

Low w ell being

Low social support


Bad housing

Not voting in the last general election

Poor access to services


Poor emotional health

Low contact w ith family and friends

Poor general health

No sense of belonging

Poor access to transport


Material deprivation

No pension w ealth

No financial services

Percent



1.6 Segmenting older people according to singular markers of risk As there were different characteristics of women and the older old within the group of older people, this section on segmentation focuses on these two groups. The age and sex profile of those experiencing each risk marker of social exclusion can be compared to that of the whole sample. The greatest disparity between men and women was with regards to pension wealth, where virtually all those who were marked by risk on pension wealth were women. In contrast men were more likely to be marked by risk in terms of contact with family or friends. Overall women experienced more risk markers than men, but there weren’t necessarily consistent patterns within domains – for example women were more likely to have poor emotional health, but were less likely to have poor general health. Figure 1-3 Proportion of those experiencing each risk marker who were women

0 10 20 30 40 50 60 70 80 90 100

Total

No pension w ealth





No qualif ications

Relative low income




Fear of crime


Low social support





Bad housing

Low w ell being

Poor general health


Percent

Base: All experiencing each risk marker


There was a stronger relationship between age and risk markers of social exclusion than there was for sex and risk markers. The older old (those aged 80+) were much more likely to be marked by risk than their younger counterparts on a wide range of measures. Over 40% of those marked by risk in terms of poor access to transport, material deprivation, lack of physical exercise and no pension wealth were aged 80 and over, compared with 16% of all older people who were aged 80 and over. There were a further seven markers of risk where the proportion of those aged 80 and over was 20% or more. Figure 1-4 Proportion of those experiencing each risk marker who were aged 80+

0 10 20 30 40 50 60 70 80

Total





No pension w ealth



Relative low income

Low social support

No f inancial services

Low w ell being

No qualif ications



Fear of crime

Poor general health


Bad housing




Percent

Base: All experiencing each disadvantage

Base: all those experiencing each risk marker of social exclusion

Understanding the risks of social exclusion in older age Multiple risk markers of social exclusion 14

2 Multiple and multidimensional risk markers of social exclusion This section focuses on the extent to which older people may be affected by more than one of marker of risk, more than one domain of exclusion, and by particular combinations of risk markers of social exclusion. 2.1 The multidimensional nature social exclusion The experience of multiple markers of risk, classed as having two or more risk markers, can have severe negative implications and consequences for quality of life, well-being and future life chances. When using ELSA to explore multidimensional social exclusion, we have focused on 16 of the 21 risk markers discussed in the previous chapter. One reason for this was to balance out coverage of the sub-domains of the B-SEM, so that no sub-domain was over represented by risk markers of social exclusion in the Latent Class Analysis. The five risk markers removed from the multidimensional analysis (housing tenure, carer, qualifications, limiting illness, and well being), are useful as explanatory characteristic variables in their own right.

Domain Subdomain Risk marker of social exclusion Resources Material and economic resources Relative low income Material deprivation No pension wealth Access to public and private services Poor access to services Poor access to transport No financial services Social resources Low social support Low contact with others Participation Culture, education and skills Poor functional literacy and numeracy skills Political and civic participation Not voted in the last general election Quality of life Health and well-being Poor self-reported general health Low participation in physical exercise Poor emotional health Living environment Bad housing No sense of belonging Crime, harm and criminalisation Fear of area after dark

In order to look at multiple markers of risk, a simple additive scale was created to measure the number of risk markers experienced by older people. Each of the 16 risk markers of social exclusion were coded as a binary 0,1 variable, and added together to get a score out of 16. When considering the additive scale, it is important to remember that the threshold of risk for each individual marker is relative and/or arbitrary.

A reliability analysis was run on the 16 risk markers, to see whether the items measured a common underlying concept (please see Annex B for further details of the analysis). A Cronbach’s alpha score of 0.5 was calculated which suggests that the 16 risk markers were tentatively measuring an

Table 2-1 Risk markers used in multidimensional social exclusion analysis


underlying concept8. The lack of a particularly high score suggests that the risk markers under investigation are also distinctly different.

The result of the reliability analysis suggests that risk of social exclusion cannot be measured in a single variable, however the additive scale does allow us to look at single versus multiple risk markers. It is in this context that we will be using the scale.

In total, 50% of older people experienced two or more of the 16 risk markers, and this experience was more prevalent among women and the older old. Women were more likely than men to experience multiple risk markers (54% compared with 44%). Older old people were much more likely than younger old to experience multiple risk markers (72% of those aged 80 and over, compared with 52% of those aged 70-79, and 41% of those aged 60-69%). Figure 2-1 Sex and age profile of those experiencing multiple risk markers

0

20

40

60

60-69 70-79 80+

Perc

ent Men

Women

Base: All respondents experiencing multiple disadvantage

Looking at family situation, 78% of those who were single, 62% of those who were widowed, and 71% of those who were divorced or separated experienced multiple risk markers. Of these non-married older people, those who lived alone were more likely than those who lived with other people, to experience two or more risk markers (72% compared with 62%). Having living siblings or children did not influence chances of experiencing two or more risk markers. 52% of city dwellers suffered multiple risk markers - those living in a city were more likely than those living in a town, village, or hamlet to be marked by risk on two or more measures. It is not just typically those in the ‘lowest’ groups who experience multiple markers of risk. 39% of those who had at least one qualification also experienced two or more risk markers, as did 26% of those on a high income, and 44% of those who owned their own home. 38% of those who were married and 46% of those who were remarried, were also likely to experience risk on at least two of the 16 risk markers.

8 A Cronbach’s alpha score of 0.7 or above would suggest that there was a distinct underlying concept.

Base: all respondents experiencing multiple risk markers of social exclusion


When considering other socio-demographic characteristics, such as income, it is difficult to examine multiple risk markers of the least well off groups with simple crosstabular analysis as by definition these groups are experiencing one marker of risk. The next step is to unpick these confounders and to investigate how these multiple risk markers overlap with each other, exploring whether some markers are particularly associated with other markers. 2.2 Which risk markers do older people experience? This section considers whether the experience of multiple risk markers is multidimensional. For example is an older person likely to experience interrelated risk markers such as health problems only, or are they just as likely to have a combination of risk markers linked to different dimensions of exclusion, for example health as well as financial problems. The analysis begins by exploring the interrelationships between markers of risk using tetrachoric correlation analysis9. Findings indicate that some of the 16 markers of risk correlate particularly highly with each other, while others do not. The highest correlation is between poor self-reported general health and not being physically active at 0.6110. The results of the tetrachoric correlation analysis, presented in Annex B, show the potential for some interesting overlapping risk markers, as some of the 16 markers from different sub-domains in the B-SEM correlate together, for example low social support and material deprivation correlate together at 0.41, and poor self-reported general health and poor access to services are correlated at 0.45. This analysis illustrates that the relationship between any two markers is seldom a strong one, suggesting that risk markers experienced by older people across the 16 markers is multidimensional, and the experience is different for different people. (Correlation matrix is in section 5.1, correlations higher than or equal to 0.20 have been highlighted). To understand further the relationships between overlapping markers of risk, we looked at those who had two or more out of a total of 16 risk markers (2156 respondents), and used Latent Class Analysis to identify particular clusters of older people. Section 2.2.1 will consider each cluster identified by the Latent Class Analysis. Older people not experiencing multiple risk markers and therefore not part of the Latent Class Analysis, are explored in part 2.3. Latent Class Analysis is a multivariate technique that is useful in exploring whether social exclusion is multidimensional. It is analogous to cluster analysis in that it is a means of identifying clusters of similar individuals who share an underlying or ‘latent’ characteristic.

9 Tetrachoric correlation analysis is suitable for pairs of dichotomous variables that assume an underlying continuous variable. 10 A correlation of 0.7 is considered high.


In each latent class identified by the analysis, a set of probabilities11 is produced for the 16 risk markers. These probabilities show for each marker, the likelihood of a person being marked by risk. These probabilities are useful when defining the latent classes identified by the analysis. (See chapter 6.2 in Annex B for a technical description of the Latent Class Analysis). 2.2.1 Latent Class Analysis findings Findings of the Latent Class Analysis suggest that among those experiencing at least two risk markers, there are five different clusters, or combinations, of risk markers experienced by older people12. The clusters were re-classified in order of mean number of risk markers experienced by those in each cluster group. Clusters were labelled according to the most prominent markers of risk, in other words when the likelihood of experiencing a risk marker was greater than 0.20. As part of this labelling process it became evident that some risk markers experienced were common among older people, and appeared in all clusters (fear of local area, bad housing, low social support, poor functional literacy and numeracy skills, and low political efficacy). As such, commonly experienced risk markers were generally not used to label clusters. Commonality of risk markers is discussed in detail in section 2.2.2.

Cluster One: Mostly lonely and unsupported 20% of older people formed a cluster that were fundamentally lonely, in that they were likely to have very infrequent contact with others, had low social support, and felt that they didn’t belong in the area that they live. Older people in this group were likely to be marked by risk to some extent in terms of access to services, as well as likely to experience some of the risk markers common to all those suffering from multiple exclusion (bad housing, poor functional literacy and numeracy skills, and low political efficacy). Unlike other clusters, individuals in this cluster of individuals were not afraid of their local area after dark. Older people assigned to this group, were very unlikely to have access problems in terms of transport or financial services. They were not physically inactive, or experienced poor general health. On average, older people in this group experienced two or three risk markers out of 16. Those in this group were primarily marked by risk on the participation domain and living environment sub-domain, in addition to being marked by risk with respect to service access. 11 Latent Class Analysis produces recruitment probabilities for each cluster. This is the probability that, for a randomly selected member of a given latent class, a given response pattern will be observed. From the recruitment probabilities, one easily calculates the a posteriori probability of an individual’s membership in each class. One may then assign the individual to the latent class with the highest a posteriori probability (modal assignment). 12 To verify findings of the Latent Class Analysis and the identification of five clusters, the analysis was repeated on wave 1 ELSA data and a similar latent class model was produced.


Figure 2-2 Cluster One: Lonely - probabilities of risk markers

0.21

0.42

0.40

0.34

0.33

0.26

0.20

0.18

0.12

0.12

0.07

0.06

0.03

0.03

0.03

0.02

0.01

0.00 1.00

Cluster probability for all respondents

Low social support

Has housing problems

Not voted

No contact w ith others

Poor literacy and numeracy skills

Feeling of not belonging


Low income



No pension w ealth

No access to private or public transport

Not physically active


Self reported poor general health

Fear of area after dark

Probability

Base: 429 respondents (20%) of those experiencing multiple risk markers of social exclusion


Cluster Two: Fear of local area Older people experiencing multiple markers of risk had a 23% chance of being in this group. Unlike cluster one, older people in this cluster were defined distinctively by one characteristic - fear of their local area after dark. Older people in this cluster were also particularly affected on other common13 risk markers (bad housing, low social support, poor functional literacy and numeracy skills, and low political efficacy). On average, older people in this cluster were likely to experience around three risk markers out of 16. This group were extremely unlikely to have a relative low income, with a probability of 0.02. Similarly those assigned to this group were very unlikely to be lonely, have poor transport access, be physically inactive, have no pension wealth or financial services, be materially deprived, or have poor self-reported general health.

13 See section 2.2.2 for a discussion of endemic risk markers.


Figure 2-3 Cluster Two: Fear of local area - probabilities of risk markers

0.23

0.30

0.27

0.23

0.20

0.17

0.15

0.14

0.13

0.06

0.05

0.04

0.04

0.04

0.03

0.02

0.00 1.00




Low social support


Not voted









No pension w ealth


Low income

Probability


Cluster Three: Low income This cluster was the largest group identified by the Latent Class Analysis, with 32% of older people assigned to this cluster. This cluster consists entirely of people who were in the lowest income group (although there were some people in the lowest income group who were not in this cluster), the other characteristics of older people in this group were that they were likely to be afraid of their local area after dark, were likely to have low social support, had poor literacy and numeracy skills as well as housing problems.


Conversely, this group were very unlikely to report poor health, or be physically inactive. They were also unlikely to have poor access to transport or poor access to financial services. Older people experiencing risk markers on two or more of the 16 ELSA markers of risk had a 28% chance of belonging to the low income cluster, as the cluster probability for all older people is 0.28. Those in this group typically experienced three out of 16 risk markers. Figure 2-4 Cluster Three: Low income - probabilities of risk markers

0.28

0.42

0.28

0.25

0.20

0.17

0.14

0.12

0.10

0.09

0.09

0.07

0.06

0.06

0.03

0.01

0.00 1.00


Low income


Low social support



Not voted

No pension w ealth










Probability of disadvantage



Cluster Four: Mostly poor health 14% of older people affected by multiple risk markers were likely to suffer multidimensional problems that spanned all three domains of the B-SEM. Older people in this group were likely to suffer health related problems in the main, and had a 50% chance of poor general health and poor emotional health. On average, older people in this cluster experienced around four of 16 risk markers. Low relative income and service access were also issues for those who were classified in this cluster, as were the endemic problems faced by the wider older population affected by multiple risk markers (bad housing, poor functional and literacy skills, fear of area after dark, low political efficacy demonstrated through not voting in the last general election, and low social support). Figure 2-5 Cluster Four: Mostly poor health - probabilities of risk markers

0.16

0.53

0.50

0.33

0.33

0.32

0.27

0.25

0.24

0.22

0.15

0.12

0.09

0.07

0.04

0.00

0.00

0.00 1.00









Low income

Not voted

Low social support




No pension w ealth



Probability

Base: 304 (14%) of those with multiple risk markers of social exclusion


Cluster Five: Mostly poor access The smallest cluster identified by the Latent Class Analysis contained 10% of older people who experienced multiple risk markers. Older people in this cluster were primarily defined by their lack of access to services and lack of access to transport. Those in this group were also likely to be physically inactive, had poor self-reported general health and emotional health, and were also likely to experience the risk markers endemic to this population. This cluster experienced more risk markers on average than any other cluster, with older people in this group likely to suffer five out of 16 risk markers. Figure 2-6 Cluster Five: Mostly poor access - probabilities of risk markers

0.12

0.66

0.58

0.43

0.41

0.41

0.36

0.33

0.32

0.23

0.21

0.20

0.18

0.12

0.10

0.08

0.06

0.00 1.00






Low social support




Low income

Not voted




No pension w ealth



Probability

Base: 205 (10%) of those experiencing multiple risk markers of social exclusion


2.2.2 Comparing clusters of multiple risk markers of social exclusion Five clusters were identified when looking at the forms of multiple risk markers experienced by older people.

• Mostly poor access (most marked by risk) - mean number of risk markers in this cluster was 5.1.

• Mostly poor health - mean number of risk markers was 3.7. • Low income - mean number of risk markers was 3.2. • Mostly lonely & unsupported - mean number of risk markers was 2.5. • Fear of local area (least marked by risk) - mean number of risk markers was 2.8.

Those in the mostly poor health cluster and mostly poor access cluster were most vulnerable to multiple risk markers. The poor health subgroup had at least a 20% chance of experiencing a risk marker of social exclusion on nine of 16 risk markers, and was therefore more likely than the low income, lonely and fear of local area clusters to experience severe multidimensional exclusion. Older people in the poor access cluster can be thought of as the most at risk of severe multiple risk markers and were likely to experience all markers to some degree. Although we have labelled clusters in terms of the markers of risk most likely to be experienced, the clusters produced by the Latent Class Analysis were by no means mutually exclusive or distinct from each other in terms of risk markers experienced in each group. Considering each of the five clusters identified by the Latent Class Analysis, those in the fear of local area cluster predominantly experienced risk markers that were common to all clusters. This cluster can be thought of as including individuals who were the least vulnerable to severe multidimensional exclusion, as the other clusters suffered further problems in addition to these common risk markers. Further analysis into the prevalence of markers of risk by cluster, demonstrates further the presence of common or “endemic” risk markers experienced by older people.


Table 2-2 Prevalence of each risk marker of social exclusion by cluster

Multiple risk marker cluster

Domain Sub-domain Risk marker Low Income

Fear of local area

Lonely Poor Health

Poor Access Total

% % % % % %

Resources Material and economic resources Low income 100 - 3 21 20 22

Material deprivation 9 6 7 - 23 4 No pension wealth 13 4 6 5 11 4

Access to public and private services Poor access to services 13 15 19 36 73 12

No access to private or public transport 5 4 2 - 76 5

No financial services 6 3 4 10 7 3 Social resources Low social support 28 26 43 13 45 17 No contact with others 10 13 34 6 14 9

Participation Culture, education and skills

Poor literacy and numeracy skills 25 29 26 28 36 16

Political and civic participation Not voted 18 20 34 23 21 14

Quality of life Health and well-being

Self reported poor general health - 5 0 67 38 7

Poor emotional health 12 14 9 63 30 11 Not physically active 3 4 2 32 51 6 Living environment Has housing problems 22 23 40 33 17 16 Feeling of not belonging 8 16 22 12 5 7

Crime, harm and criminalisation Fear of area after dark 40 100 - 25 42 28

Base 701 527 429 304 205 2166 Having poor functional and literacy skills is one risk marker that was consistently present across all five clusters with a similar proportion of older people experiencing risk markers. Bad housing, fear of local area after dark, low social support, and low political efficacy as demonstrated by not voting in the last general election, were common in at least four out of five of the clusters. The presence of these endemic risk markers is an important caveat to the multidimensional analysis that needs to be kept in mind when interpreting the findings of the multidimensional analysis. These common or endemic risk markers originate from different domains and sub-domains of the B-SEM. This means that older people who experience two or more markers of risk, are by definition, experiencing a form of multidimensional exclusion. Despite some commonality between clusters, there is also a clear distinction between other clusters. For example, 100% of older people in the low income cluster had a relative low income,


whereas no older people assigned to the fear of local area cluster were income poor. 76% of older people in the poor access cluster were marked by risk in terms of access to transport and 73% were marked by risk with respect to accessing services – a proportion much higher than in other clusters. In the poor health cluster, roughly twice as many older people reported poor general health (67%) and poor emotional health (63%) than in the other clusters. To further investigate the different forms of risk markers experienced, this section will now focus on each of the five clusters identified in the Latent Class Analysis, considering the markers of risk that define the individuals in these groups, and the socio-demographic characteristics that explain them.


2.3 How do clusters of multiple risk markers of social exclusion vary? Those who experience multiple risk markers of social exclusion form an especially deprived group. The five clusters identified by the Latent Class Analysis go some way to explaining the different types of multidimensional exclusion experienced by older people. The remainder of this section examines multidimensional social exclusion as defined by the five cluster groups, in the context of a lack of risk markers and the experience of singular markers of risk.14

Demographic characteristics Men were more likely than women to experience no risk markers (26% compared with 20%) and also less likely to experience singular markers of risk (30% compared with 26%). Thus women were more likely than men to experience multidimensional social exclusion, and this difference was most pronounced in clusters related to low income (20% compared with 12%) and access to services (6% compared with 4%). Figure 2-7 Multidimensional social exclusion clusters by sex

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely

Multi: Fear of local area

Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

Men

Women


Age showed a stronger relationship with risk markers than sex, with the older old being more likely to experience multidimensional social exclusion – 72% of those aged 80 and over experienced multidimensional social exclusion compared with 41% of 60-69 year olds. As with the difference between men and women, the difference in any experience of multidimensional social exclusion

14 Ethnicity is not used in this section as an explanatory variable, due to the small sample of non-white older people in the sample. Other explanatory variables that did not have a significant effect, have been omitted from this section. However tables for these are included in chapter 5.2.


experienced by the older old was explained by greater prevalence of the multidimensional clusters of low income (28% 80+; 10% 60-69s) and poor access (16% 80+; 2% 60-69s). Figure 2-8 Multidimensional social exclusion by age

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

60-69

70-79

80+



Generally, those who were married were less likely to experience multidimensional social exclusion (39%) than those who were not married (70%). Those who were not married were more likely to be classified in most multidimensional clusters apart from poor health, and in particular were more likely to be in the low income cluster (28% compared with 11%). As marital status is related to age, it may be that age confounds this analysis, but the same patterns of differences were seen between married and non-married older people when the analysis was restricted to 60-69 year olds. Figure 2-9 Multidimensional social exclusion by martial status

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

Married

Non-married



Another factor that may affect the experience of multiple markers of risk is whether older people live alone or with someone else. As living arrangements would be confounded by whether people are married or not, this analysis was restricted to those who were not married. Among unmarried older people those living alone were more likely to experience multidimensional social exclusion (73% compared to 62%), and the greatest difference was in being in the low income (29% compared to 20%) and poor access clusters (12% compared to 7%). Figure 2-10 Multidimensional social exclusion by living arrangements (not married)

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

Lives alone

Lives w ithothers

Base: All non-married respondents

There was no relationship between multidimensional social exclusion and whether older people had living siblings.


Older people without qualifications were more likely than those with qualifications to experience multidimensional exclusion (62% compared with 39%). Those without qualifications were more likely to be in clusters related to low income, poor health and poor access to services, but were no more likely to belong to loneliness or fear of the local area clusters. Although age was related to whether people had qualifications (the older old were less likely to possess them), this did not confound the analysis and similar patterns were seen when the analysis was restricted to just 60-69 year olds. Figure 2-11 Multidimensional social exclusion by whether has qualifications

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

Qualif ications

None



Home and area characteristics Home ownership was strongly related to being marked by risk, with those who rented being much more likely to experience multidimensional social exclusion (78%) than those who owned their home (44%). Those who rented were more likely to be in every cluster of multiple risk markers apart from the loneliness one. Figure 2-12 Multidimensional social exclusion by whether owns own home

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

Ow n

Rent



Being marked by risk was related to population density with those in cities (52%) more likely to experience multidimensional social exclusion than those in towns (47%) and villages (39%). However, when considering the relationship between cluster membership and population density there was no significant relationship15. Figure 2-13 Multiple markers of risk by locality

0 5 10 15 20 25 30 35 40 45 50 55

All experiencing multiple disadvantage

City dw eller

Tow n dw eller

Village or hamlet dw eller

Percent

All experiencing multiple disadvantage

City dw eller

Tow n dw eller

Village or hamlet dw eller

Base: All experiencing multiple risk markers of social exclusion

15 Crosstabulation analysis of population density by multiple risk marker cluster membership in the context of singular risk markers and lack of risk marker is presented in Annex A, Section 5.3 at the end of this chapter.


Given that all older people in one of the multidimensional social exclusion clusters had a low income, this confounds further analysis by personal income. However it is still possible to look at the relationship between income and risk markers, by using the Index of Multiple Deprivation (IMD) quintiles which measure the deprivation of an area (defined at the ward level). Personal income is highly correlated with IMD index of an area. It should be noted that by definition we would expect people living in an area with a high IMD index to be more likely to experience multidimensional social exclusion themselves, as IMD is itself a measure of multiple risk markers. The deprivation level of an area was related to whether people experience multidimensional social exclusion, with the exception of the loneliness cluster where there was no difference in membership of the cluster between those in highest and lowest area quintiles of IMD. However, this lack of a relationship is partly explained by the fact that the IMD does not use any indicators of social support in its definition. Figure 2-14 Multidimensional social exclusion by Index of Multiple Deprivation Quintile

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

Leastdeprived

Mostdeprived



Health factors Limiting longstanding illness and low well-being are to some extent confounded in this analysis as one of the clusters is mainly about poor health. Nevertheless, it is still useful to consider these as explanatory variables of social exclusion for the older population. Older people with a limiting longstanding illness were more likely to experience multiple risk markers than those who did not have such an illness (64% compared with 42% respectively). Those with a limiting longstanding illness were more likely to be in clusters related to poor access to services, poor health and fear of the local area, but were no more likely to belong to loneliness or poor access to services clusters. Figure 2-15 Multidimensional social exclusion by whether have limiting longstanding illness

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

LLI

None



76% of older people experiencing low well-being experience multiple risk markers. Low well being is to some extent associated with all clusters, with those in the poor health, poor access and low income clusters most likely to have a low well being. Figure 2-16 Multidimensional social exclusion by wellbeing

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

High

Low



Whether older people care for someone is strongly related to age and this would potentially confound the relationship between caring and multidimensional social exclusion - therefore the analysis was restricted to those aged 60-69. When restricted to this age group there was no relationship between caring and multidimensional social exclusion. Figure 2-17 Multidimensional social exclusion by caring

0 5 10 15 20 25 30 35

No risk marker

One risk marker

Multi: Lonely


Multi: Low income

Multi: Poor Health

Multi: Poor Access

Percent

Carer

Not

Base: All respondents aged 60-69

Summary To summarise these findings, those who were in the cluster most marked by risk, defined as those experiencing mostly poor access problems, were likely to be aged 80 and older, had no qualifications, were living alone, and had a low well-being. Having no qualifications and a low wellbeing, was also associated with membership in the mostly poor health cluster, as was having a limiting longstanding illness, The low income cluster also included older people who had no qualifications and a low well-being. Older people in this cluster were also likely to live alone, and were particularly likely to be unmarried when compared with other clusters. Renting accommodation as opposed to owning or buying a home, and being in the lowest quintile of the Index of Multiple Deprivation, was associated with older people in all clusters, with the exception of those who were mostly experiencing loneliness and low social support. Being a carer was not a risk marker of any of the five forms of multidimensional social exclusion identified in the analysis. Nor was population density when considering the relationship between an older person living in a city, town, or village/hamlet, with multidimensional social exclusion.

Understanding the risks of social exclusion in older age Social exclusion risk marker dynamics 38

3 Social exclusion risk marker dynamics 3.1 Measuring the dynamics of social exclusion Multidimensional social exclusion is a process that can affect people over a period of many years, where they may get into a vicious circle of risk markers and find it hard to escape their circumstances. Levitas et al (2007) called this process deep exclusion. Chapters 1 and 2 examined a snapshot of exclusion in one particular year using ELSA data. This chapter will explore the dynamics of multiple exclusion, considering duration and recurrence of exclusion, as well as the events or ‘triggers’ that cause it. In order to be able to look at exclusion over a number of years, we needed to use BHPS data rather than ELSA (which had 3 waves of data at the time of this analysis). The BHPS risk markers used in the investigation of social exclusion dynamics, were similar in terms of their derivation to the ELSA risk markers used in the study of multidimensional social exclusion. As with ELSA, the BHPS was not designed to measure social exclusion, and its longitudinal nature means that tracking trends over time is restricted to those questions asked in every wave. Thus, the risk markers derived in the BHPS data are fewer than in the ELSA data as shown in the table below.

Table 3-1 B-SEM risk markers of social exclusion

Domain Sub-domain Risk marker of social exclusion ELSA BHPS Resources Relative low income

Material and economic resources Receipt of out of work benefits

Material deprivation Fuel Poverty Not owning own home No pension wealth No wealth and savings Debt Subjective poverty Poor access to services

Access to public and private services Poor access to transport

No financial services No private services Poor utilities Social resources Institutionalisation Low social support Low contact with others


Domain Sub-domain Risk marker of social exclusion ELSA BHPS Participation Economic participation Cares for another person In employment Living in a workless household Undertaking unpaid work Poor quality of working life Social Participation No participation in positive activities Culture, education and skills Poor functional literacy and numeracy School absences and exclusions Not doing cultural, leisure activities

Not participating in cultural and Sporting activities

No internet access No qualifications Political and civic participation Not voted in the last general election Quality of life Health and well-being Poor self-reported general health Low participation in physical exercise Obesity Limiting longstanding illness Poor emotional health Low well being Low personal efficacy Long periods on benefits Smoking Drug use Living environment Bad housing Homelessness Poor neighbourhood safety Low area satisfaction No sense of belonging Limited access to space Fear of area after dark

Crime, harm and criminalisation Experience of crime

Hospital admissions Domestic violence Fear of crime Bullying Discrimination Committed a crime Imprisonment Has an Anti Social Behaviour Order

The BHPS risk markers selected for analysis cover eight of the ten themes of the B-SEM. As with previous analyses, some of these markers were used as characteristic variables and were used in the formulation of the Latent Class model. In investigating the dynamics of social exclusion, we have repeated the analysis techniques that were used when exploring the multidimensional exclusion in the previous chapter. Latent Class Analysis was used to identify different combinations of multiple risk markers. The movement of these subgroups of older people was explored over time.


The difficulty in analysing different markers of risk from several consecutive survey years, is that questionnaire structure is not always the same year after year. A balance needs to be found between selecting the optimal number of risk markers for the analysis and the availability of these markers over time. Table 3-2 illustrates the risk markers chosen for the Latent Class Analysis stage of the longitudinal analysis. These cover eight of the ten themes of the B-SEM.

Table 3-2 Risk markers used in longitudinal multidimensional analysis

Wave 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Year 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 Low relative income


No transport access

Low social support M M M M M M M

Low contact with others

No qualifications

Poor general health M


Bad housing M M M M M

Area satisfaction indicates availability of risk marker in a wave.

M indicates missing risk marker in a wave.

Due to the changing nature of the questionnaire content wave on wave, the social support risk marker of social exclusion was not available in every survey year. Self-reported general health was also not measured in one wave. Because of this, social support and general poor health were derived for those waves in which these questions were not asked16. It could be argued that more markers of risk could have been derived using a similar methodology. However, it did not seem sensible to derive risk markers when there was substantial missing data due to changing questionnaire structure between survey years. In longitudinal analysis of this type, there needs to be a balance between guessing what a person might have said in a year, for which a question was missing, and excluding this risk marker altogether. 3.2 Profile of the balanced panel sample Our analysis focuses on a panel sample of 987 individuals who were aged 60 and over in 1997, and were interviewed annually from this year until 2005. In the first wave of the analysis (1997) 42% of the balanced panel sample were men and 58% women. 59% of older people were aged 60-69, compared with 36% aged 70-79, and 8% aged 80 and older. The sex and age profile of the balanced panel sample is markedly different from that seen in the ELSA population which reflected that of the general older population. Age distribution

16 Imputed risk markers were derived from complete data at preceding and following waves. Prevalence of

risk on derived markers was therefore in line with the prevalence of risk markers in the preceding and following wave.


among older people in the balanced panel is particularly different, and this may be a result of attrition of older old people aged 70 and over the course of the survey which started in 199117. Further analysis of the socio-demographic characteristics of the panel sample is presented in section 5.4 of Annex A. Table 3-3 Prevalence of risk markers experienced by panel sample in 1997

Material and economic resources Risk Marker Prevalence Govt Indicator Definition

Relative low income 26% PSA17 Household income is less than 60 percent of contemporary median household income as measured by BHPS, before housing costs are deducted, and equivalised using the McClements Scale.

Material deprivation 9% OA3 Possess three or fewer of the following: central heating, freezer or fridge freezer, washing machine, microwave oven, telephone, home computer.

Not owning own home 24% Living in accommodation which is shared ownership, rented, or something other than owned or buying with the help of a mortgage.

Access to public and private services Risk Marker Prevalence Govt Indicator Definition

Poor access to transport 30% OA8 No access to a car and or van for private use.

Social resources Risk Marker Prevalence Govt Indicator Definition

Low social support 38% Scored one or more on the derived 5 point social support scale

Low contact with others 51% OA6 Don’t meet or talk to people at least once in a week

Economic participation Risk Marker Prevalence Govt Indicator Definition

Living in a workless household

70% Living in household where no-one works

Culture, education and skills Risk Marker Prevalence Govt Indicator Definition

No qualifications 48% Having no qualifications

Health and well-being Risk Marker Prevalence Govt Indicator Definition

Poor self-reported general health

9% Self-reported general health is ‘Poor’

17 As the profile of the panel sample used in this study is not representative of older people in the general population, findings presented in this chapter reflect societal change rather than a change in the general population of older people between 1997 and 2005.


Poor emotional health 13% Score at least 4 on the GHQ12 scale

Living environment Risk Marker Prevalence Govt Indicator Definition

Bad housing 26% Have at least 1 of the following housing problems: Shortage of space; Too dark; Rising damp; Water from roof etc.; Condensation; General rot and decay; Too cold in winter.

Low area satisfaction 6% Dislikes neighbourhood

Note prevalence of risk marker , for years 1997 to 2005, is presented in section 5.5 of Annex A. 3.3 The behaviour of multidimensional social exclusion over time As with the cross-sectional ELSA analysis, a Latent Class Analysis was run on BHPS data for all those experiencing multiple markers of risk, to form clusters of multidimensional social exclusion.18 Then additional groups of those experiencing no risk markers or one risk marker were added after the Latent Class Analysis. It is difficult to compare the BHPS clusters to the ELSA clusters because different variables were used in the different latent class analyses, and because there were 3 BHPS clusters and 5 ELSA clusters. However, looking at the common variables used in each analysis suggests that the ELSA and BHPS clusters map onto each other as shown in Figure 3-1. Figure 3-1 Relationship between ELSA and BHPS multidimensional clusters

18 Different Latent Class Analyses were run on data from 1997, 2001, 2003 and 2004 to check whether the

cluster solutions were stable over time. These years were the four survey waves for which all indicators were available.

Low income

Poor health

Poor access

Lonely

Fear of area

Health & low social support

Low social support & lonely

Low income & transport

BHPS

ELSAClusters

Low income

Poor health

Poor access

Lonely

Fear of area

Health & low social support

Low social support & lonely

Low income & transport

BHPS

ELSAClusters


The same endemic risk markers were identified in all clusters produced by the model (low educational attainment, low social support, bad housing) and the cluster definitions themselves were similar, although the BHPS model suggested a three cluster model was the most appropriate fit for the data, compared to a five-cluster model identified in the ELSA data. The three clusters of older people identified, can be thought of as a somewhat compressed translation of the clusters identified in the ELSA data. The three BHPS clusters followed a somewhat hierarchical nature, in that cluster one could be thought of as the subgroup least marked by risk; cluster three could be thought of as the most vulnerable to multiple exclusion; and cluster two was somewhere in the middle. Each cluster, reflects by nature, multidimensional social exclusion, in the sense that risk markers experienced in each of the three spans more than one domain of the B-SEM. In chapter two, we explored the multidimensional social exclusion experienced by older people. Using the BHPS we are able to look at the length of time that older people experience exclusion, and whether or not their experience of exclusion is recurrent. A useful output of Latent Class Analyses is the probability scores discussed in detail in chapter 4.2 of the technical annex. New members of a sample can be assigned to an already defined cluster structure using probability scores, or can be used to impose the cluster structure on a new wave of panel data. It is possible therefore, to explore the behaviour of different clusters of individuals longitudinally. In this analysis, the duration of particular forms of social exclusion and its recurrence, were explored longitudinally between 1997 and 2005, through the application of assigning a defined cluster structure identified in wave 1997 of the panel data, to the following consecutive eight waves.19 The rest of this chapter examines each multidimensional cluster in turn20.

19 The cluster structure applied to all waves was checked for consistency against other survey years, and

full details of this modelling approach can be found in Annex B. 20 Annex A, sections 5.5, 5.6, 5.7, 5.8, and 5.9 present analysis on multiple risk markers and cluster

dynamics.


Low social support and loneliness cluster This cluster consisted of 40% of those experiencing multiple social exclusion in 1997, and the mean number of risk markers experienced by this cluster was 2.6.21 Over the nine waves, men were more likely than women to be in this group (20% compared with 17%) unlike the other two clusters where women were predominant. Figure 3-2 Cluster One: Low social support and loneliness cluster - probabilities of risk markers

0.23

0.81

0.73

0.58

0.53

0.24

0.15

0.11

0.03

0.01

0.00

0.00 1.00

Cluster probability for allrespondents

Low social support

No qualif ication

Limited contact

Bad housing

No private transport

Low area satisfaction

Low income


Poor mental health

Poor health

Probability

Base: 261 (46%) of those experiencing multiple risk markers of social exclusion at wave 1997

Over the 9 waves of BHPS, 83% were in the low social support and loneliness cluster on at least one occasion. Among those who were in the cluster at least once, the mean number of occurrences in the cluster was 4.2. No-one was in this cluster for all 9 waves, but 18% were in the cluster on more than two-thirds of occasions (at least 7 out of 9 years).

21 Note that the mean number of risk markers for BHPS multidimensional clusters was lower than for ELSA

clusters because there were more risk markers of social exclusion in the ELSA analysis.


Figure 3-3 Duration in low social support and loneliness cluster

0 5 10 15 20 25 30

1

2

3

4

5

6

7

8

9

Occ

asio

ns in

Clu

ster

Percent

Base: All in low social support and lonely cluster on at least one occasion

In terms of moving between clusters between waves of BHPS, of the people who were in the low social support and loneliness cluster at a wave, 44% were likely to still be in the same cluster at the subsequent wave, 39% were likely to move to one of the other multi-risk marker clusters and 17% were likely to move out of multidimensional social exclusion (15% to singular risk marker and 1% to no risk marker22). There was a similar pattern of movement seen in the low social support and loneliness group between one year and the next – 46% had been in the same cluster, 38% had been in another multi-risk marker cluster and 17% had not been in a multi-risk marker cluster. There is clearly some churn in membership of the cluster over time, although on balance people in this cluster who move out of it were more likely to move to a different situation of multiple risk 22 Figures do not add up to 17 due to rounding.

Table 3-4 Duration in low social support & loneliness cluster

Base: All in low social support and lonely cluster on at least one occasion BHPS

Total Years % 1-3 404-6 417-9 18Base 822


markers than to move out of situation of multiple risk. Given that this cluster has the lowest mean number of risk markers out of all the multi-risk marker clusters, then people in this cluster at one wave were more likely to see their situation worsening rather than improving at the next wave. Figure 3-4 Persistence in low social support and loneliness

cluster

t-1

t

t+1

None

Singular

Other multi

Low SocialSupport &Loneliness


Low income and transport cluster This cluster consisted of 40% of the multi-risk marker sample in 1997, and the overall mean number of risk markers in the cluster was 3.3.23 Older people in this cluster were likely to be women (31% women, compared with 21% men). This cluster was most likely to apply to older old people than the other clusters (40% aged 80+ compared with 30% aged 70-79, and 18% aged 60-69). Figure 3-5 Cluster Two: Low income and transport cluster - probabilities of risk markers

0.55

0.74

0.72

0.62

0.43

0.39

0.36

0.24

0.11

0.10

0.06

0.00 1.00


No qualif ication


Low income

Limited contact

Low social support

Bad housing


Poor health

Poor mental health


Probability


Over the 9 waves of BHPS, 64% were in the low income and transport cluster on at least one occasion. Among those who were in the cluster at least once, the mean number of occurrences in the cluster was 4.5. 10% of those who were in the cluster at least once were in this cluster for all 9 waves, and 28% were in the cluster on more than two-thirds of occasions (at least 7 out of 9 years).




Figure 3-6 Duration in low income and transport cluster

0 5 10 15 20 25 30

1

2

3

4

5

6

7

8

9

Occ

asio

ns in

clu

ster

Percent

Base: All in low income and transport cluster on at least one occasion

In terms of moving between clusters between waves of BHPS, of the people who were in the low income and transport cluster at a wave, 68% were likely to still be in the same cluster at the subsequent wave, 26% were likely to move to one of the other multi-risk marker clusters and 7% were likely to move out of multidimensional social exclusion (almost all of these to singular risk marker). There was a similar pattern of movement looking back at which clusters people in the low income and transport cluster were in at the previous wave – 67% had been in the same cluster, 33% had been in another multi-risk marker cluster and 8% had not been in a multi-risk marker cluster. There was less movement in and out of this cluster over time than there was for the other clusters. Again on balance people in this cluster who move out of it were more likely to move to a different combination of multiple markers of risk than to move out of an experience of multiple markers.

Table 3-5 Duration in low income and transport cluster

Base: All in low income and transport cluster on at least one occasion BHPS

Total Years % 1-3 434-6 297-9 28Base 631


Figure 3-7 Persistence in low income and transport cluster

t-1

t

t+1

None

Singular

Other multi

Low Income &Transport


Health and loneliness cluster This cluster consisted of 20% of the multi-risk marker sample in 1997, and the overall mean number of risk markers in the cluster was 3.4.24 On average, women were more predominant than men in this cluster (18% compared with 13%). People in this group were no more likely to be older old than to be younger old. Figure 3-8 Cluster Three: Health and loneliness cluster - probabilities of risk markers

0.22

0.68

0.63

0.57

0.47

0.39

0.33

0.21

0.18

0.12

0.09

0.00 1.00


Poor mental health

Limited contact

Low social support

No qualif ication

Poor health

Bad housing

Low income




Probability


Over the 9 waves of BHPS, 65% were in the health and loneliness cluster on at least one occasion. Among those who were in the cluster at least once, the mean number of occurrences in the cluster was 3.0. No-one was in this cluster for all 9 waves, but 7% of those who were in the cluster on at least one occasion were in the cluster on more than two-thirds of occasions (at least 7 out of 9 years).




Figure 3-9 Duration in health and loneliness cluster

0 5 10 15 20 25 30

1

2

3

4

5

6

7

8

9

Occ

asio

ns in

Clu

ster

Percent

Base: All in low social support and lonely cluster on at least one occasion

In terms of moving between clusters between waves of BHPS, of the people who were in the health and loneliness cluster at a wave, 54% were likely to still be in the same cluster at the subsequent wave, 28% were likely to move to one of the other multi-risk marker clusters and 18% were likely to move out of multidimensional social exclusion (16% to singular risk marker and 2% to no risk markers). Looking back at which clusters people in the health and loneliness cluster were in at the previous wave – 51% had been in the same cluster, 30% had been in another multi-risk marker cluster and 19% had not been in a multi-risk marker cluster. Although this was the cluster with the highest mean number of risk markers, people who moved out of the cluster were more likely to move out of the experience of multiple risk markers than to move to a different combination of multiple risk markers when compared to other clusters.

Table 3-6 Duration in health and loneliness cluster

Base: All in health and loneliness cluster on at least one occasion BHPS

Total Years % 1-3 664-6 277-9 7Base 642


Figure 3-10 Persistence in health and loneliness cluster

t-1

t

t+1

None

Singular

Other multi

Health andLoneliness

3.4 Comparing combinations of multiple risk markers of social exclusion over time One of the problems with using BHPS data to segment those experiencing multiple risk markers is that there were fewer clusters than the ELSA segmentation (3 compared with 5), which will be driven by the fact that there were 16 risk markers used in the ELSA Latent Class analysis compared with 10 in the BHPS analysis. It should also be noted that low social support and lack of contact with others, were drivers of both the low social support and loneliness cluster and the health and loneliness cluster, which reduces the difference between these two clusters. The 3 BHPS clusters were also less distinct than the ELSA clusters – in terms of mean number of markers of risk the BHPS clusters ranged from 2.6-3.4 mean markers of risk, whereas the range for ELSA clusters was twice as large 1.6-3.2.25 Thus it should be borne in mind that the BHPS analysis examining duration in clusters and movement between clusters may simplify some of the relationships which could be drawn out if there were more waves of ELSA to be able to carry out an equivalent analysis. The severity of the risk of social exclusion in each of the BHPS clusters was inversely related to their size. The largest cluster was low social support and loneliness which comprised 46% of those experiencing multiple risk markers at a wave and older people in this cluster experienced a mean of 2.6 markers of risk. The equivalent figures for the low income and transport were 32% of the

25 Note that the ELSA range has been scaled to be comparable. The actual range based on 16 indicators

was 2.5-5.1.


multi-risk marker population and mean of 3.2 markers of risk, and for the health and loneliness cluster were 22% and 3.4 markers of risk. Although at any one wave the low income and transport cluster was larger than the health and loneliness cluster, over all 9 waves used in the analysis, there were similar proportions of the total sample who were members of each of clusters on at least one occasion (64% and 65% respectively). In comparison, 83% of the population of older people was in the low social support and loneliness cluster on at least one of the 9 waves. In terms of long-term cluster membership, of those who had at least one occasion in a cluster, 28% of those in the income and transport cluster were in this cluster on at least 7 of the 9 waves. Equivalent figures for the other clusters were 18% for low social support and loneliness and 7% for health and loneliness. These apparent anomalies can be explained by the transition between clusters. Of those in the low income and transport cluster at one wave, 68% were likely to remain in that cluster at the subsequent wave, whereas the equivalent likelihood for the health and loneliness cluster was 54% and for the low social support and loneliness it was 44%. Those in the low income and transport cluster were less likely to move out of a multi-risk marker situation (7%) than those in the health and loneliness or low social support and loneliness clusters (18% and 17% respectively). Figure 3-11 Movement to different clusters at t+1

None Singular Low SocialSupport &Loneliness

Low Income& Transport

Health &Loneliness

None

Singular

Multiple (other)

Multiple (samecluster)

We would expect people in the cluster with the highest mean number of markers of risk to be the least likely to move out of multiple risk markers, simply as a result of having to come out of more situations of risk markers. This appears not to be the case with this analysis. However, this may be


an artefact of the extent that the BHPS analysis managed to discriminate between clusters, as we have seen that the ELSA analysis produced clusters with a wider range of mean marker of risk. The mean number of risk markers experienced by the low income and transport cluster and the health and loneliness cluster were in fact very similar (3.2 compared with 3.4), so there must be another explanation for why people are less likely to escape from the low income and transport cluster. Rather than looking at whether people remain in the same cluster in the next wave, it is possible to look at movement for each individual risk marker, to see the extent of movement. The no qualifications and poor transport risk markers were the most persistent (and low income was also relatively persistent), and the health and mental health risk markers were the least persistent. This explains why there was more movement out of the health and loneliness cluster than out of the income and transport cluster. Figure 3-12 Persistence of individual risk markers at t+1

0 20 40 60 80 100

No qualif ications

Poor transport

Low social support

Low income


Satisfaction w ith area

Poor contact

Housing problems

Poor mental health

Poor health

Percent

Base: All experiencing individual disadvantage

It should be noted that some of the movement in these risk markers (and hence between clusters) may be due to the underlying reliability of the markers used (ie would people give the same answer at different points in time if nothing had changed). The measurement of lack of qualifications is completely factual, unlikely to change much with this cohort of older people and depending on the complexity of the question, the measurement should be reasonably reliable. In contrast measurements of health status (particularly just using one question about general health) are much more likely to be affected by short term situations and people’s mood, and hence may be intrinsically less stable than other measurements. This needs to be taken into account when interpreting the results and drawing up policy recommendations. It may not be the case that in reality people are more likely to move in and out of a poor health and loneliness cluster, it may instead be an artefact of the reliability of the risk markers used.

Base: all experiencing individual risk markers of social exclusion


3.5 Triggers of becoming marked by risk There are a number of event triggers which may cause people to become marked by risk. As we have longitudinal data, we can define changes in status such as becoming retired since the previous wave and look at their relationship with a change in risk marker and a change in cluster membership. This analysis is therefore based on events, which gives us a much larger total sample size (8883) as each transition of a sample member from one wave to the next is an event. There were 7 different triggers that could be defined using BHPS data. Note that for most people each event would usually only happen once during the transition between the 9 waves of BHPS used. The triggers defined were: • becoming retired (4% of events); • starting to draw a pension (3% of events); • moving house (3% of events); • having an accident (8% of events); • becoming widowed (2% of events); • getting divorced (<0.5% of events); • getting married (<0.5% of events). It should be borne in mind that even with a large sample of events, getting divorced or married were still extremely rare occurrences and therefore it is difficult for these events to show a significant relationship with changes in risk marker. Before examining the potential impact of these triggers on cluster membership, we first need to look at whether they are related to any changes in risk marker. Thus, for example, becoming retired was related to becoming marked by risk on the low income and poor mental health risk markers. Table 3-7 Becoming marked by risk by trigger event

Trigger event

Change in risk markers Retired Pension

Move house Accident Widowed Divorced Married

Low social support Low income

Material deprivation Poor transport

Housing problems Poor contact

No qualifications Poor health

Poor mental health Satisfaction with area significant relationship


The next step is to examine whether any of these triggers are related to a change from a non-multiple risk marker situation to a multi-risk marker one. Getting divorced between one year and the next, or becoming widowed, were the only events that were significantly related to a change in situation from experiencing none or one risk marker of social exclusion to experiencing multiple risk markers. Becoming widowed was related to becoming marked by risk on 4 measures (low social support, low income, poor transport and poor mental health), all of which were strong drivers of the membership of each of the cluster groups and the end result was that becoming widowed was related to a net move into the health and loneliness cluster. The health and loneliness cluster was the one with the highest mean number of markers of risk, so perhaps in this instance it is the number of risk markers experienced that is driving the relationship between becoming widowed and membership of this cluster. Getting divorced was related to becoming marked by risk on two measures (material deprivation and housing problems), neither of which were strong drivers of cluster membership. This probably explains why although getting divorced was related to becoming at risk of multiple markers, there was no relationship between getting divorced and becoming a member of any particular multi-risk marker cluster group. Although having an accident was related to becoming marked by risk on 3 measures (poor transport, poor health and poor mental health), it was not related to becoming at risk of multiple markers. This may be due to the fact that older people who were already in the poor health and loneliness cluster were more likely to have had an accident.

Understanding the risks of social exclusion in older age Policy directions 57

4 Policy directions This study has identified new information about the different types of social exclusion older people face. The obvious policy direction recommended by this report is a focus on the 50% of individuals aged 60 and over who experience multiple risk markers of social exclusion. At a basic population level, women and the oldest old were identified as most at risk of experiencing multiple risk markers. A policy focus on social exclusion should consider these two groups, particularly as the fastest growing age group in the UK is those aged 80 and over (of which there are currently 2.7 million in the population), and the next few years will see women born in the immediate World War Two baby boom reaching pensionable age. However, a general issue for policy makers is the sheer number of older people in these two ‘at risk’ groups. The size and representativeness of the 2004-5 ELSA data used in the cross-sectional analysis in this research led to the identification of five different combinations of multiple risk markers. Individuals facing one of these combinations of multiple risk markers were categorised by the most predominant problem they experienced – access problems in terms of services and transport, health problems, low income problems, loneliness and support problems, or fear of local area problems. While using these typologies may be useful in developing focused and singular strands of policy, it is important to note that older people in these groups experienced multiple problems that were not necessarily restricted to the somewhat singular definition given above. A key finding of this study was that all those experiencing multiple risk markers shared common problems. Having poor functional and literacy skills, living in bad housing, having low social support, and having a low political efficacy were endemic among all older people marked by risk, and as such must be considered in policy. Considering the experience of multiple risk markers over time can inform social exclusion policy. Evidence in this study suggests that older people who have a combination of low income and transport problems are likely to experience multiple risk markers for the longest period of time. Older people experiencing a combination of loneliness and health problems, or low social support problems, experience multiple risk markers for a shorter duration. However, this does not mean that these groups should be seen as less important in the eyes of policy makers. Above all else, the research has shown that the combinations of multiple risk markers experienced by older people are complex and interrelated. Key targets for action suggested by the work include improving access to services and transport, poor general and mental health, low income, loneliness and low social support. In developing a comprehensive strategy to tackle exclusion, an understanding of the socio-characteristics of people experiencing multiple risk markers must be considered alongside the types of exclusion identified. Typically those at risk of experiencing any of the five combinations of multiple risk markers are older people, who live alone, are single, physically inactive, have a limiting illness, have poor health in general, and have a low well-being. This research highlights the need for policy makers to take a ‘joined up’ approach in tackling the issue of older people experiencing multiple risk markers of social exclusion. Close co-ordination is needed between Departments, Local Authorities and third sector organisations in delivering strategies to keep older people in touch with services and society more generally.

Understanding the risks of social exclusion in older age References 58

REFERENCES

Age Concern (2009) Social Inclusion and Communities, [online] www.ageconcern.org.uk/AgeConcern/ social_inclusion_what.asp : Age Concern UK

Banks, J., Breeze, E., Lessof, C., Nazroo, N. (eds) (2006) Retirement, health and relationships of the older population in England: The 2004 English Longitudinal Study of Ageing. Institute of Fiscal Studies, London. Barnes, M., Heady, C., Millar, J., Middleton, S. Tsakloglou, P. and Room, G. (2002), Poverty and social exclusion in Europe, Cheltenham, Edward Elgar. Barnes, M. (2005) Social exclusion in Great Britain: An empirical investigation and comparison with other European Union member states, Aldershot: Ashgate Publishing Limited. Barnes, M., Blom, A., Cox, K., Lessof, C. (2006) The social exclusion of older people: Evidence from the first wave of the English Longitudinal Study of Ageing (ELSA), London: Social Exclusion Unit, Office of the Deputy Prime Minister. Berthoud, R. (2003) Multiple disadvantage in employment – a quantitative analysis, York: Joseph Rowntree Foundation. Burchardt, T., Le Grand, J., Piachaud, D., (1999) ‘Social Exclusion in Britain 1991-1995’, Social Policy and Administration, Vol. 33, No. 3, pp. 227-244. Department for Work and Pensions (2005) Opportunity Age: Meeting the challenges of ageing in the 21st century, HM Government. Department for Work and Pensions (2007) Opportunity for All: Indicators update, London: HMSO. Gordon, D., Townsend, P., Levitas, R., Pantazis, C., Payne, S., Patsois, D., Middleton, S., Ashworth, K., and Adelman, L. (2000) Poverty and social exclusion in Britain, York: Joseph Rowntree Foundation. Gordon, D., Fahmy, E. (2008) Understanding social exclusion across the life course: Working age, Cabinet Office. Vermunt, J. K., and Magidson, J. (2005) Latent GOLD 4.0 User’s Guide, Belmont, Massachusetts: Statistical Innovations Inc. Lessof, C. and Jowell, R. (1999) Measuring Social Exclusion, Statistics Users Council Conference Maslow, A.H. (1968) Towards a Psychology of Being. Second Edition. Princeton New Jersey: Van Nostrand. Levitas, R., Pantazis, C., Fahmy, E., Gordon, D., Lloyd, E. and Patsios, D. (2007) The Multidimensional Analysis of Social Exclusion, London: Social Exclusion Task Force. Nolan, B. and Whelan, C.T. (1996) “Measuring poverty using income and deprivation indicators:

Understanding the risks of social exclusion in older age References 59

Alternative approaches”, Journal of European Social Policy, Vol. 6, pp. 225-240. Pantazis, C., Gordon, D. and Levitas, R. (eds) (2006) Poverty and social exclusion in Britain: The Millennium Survey, Bristol: The Policy Press. Room, G. (1998) “Social exclusion, solidarity and the challenge of globalisation”, International Journal of Social Welfare, Vol.8, No.3, pp.166-74. Scharf, T., Phillipson, C., Smith, A.E. and Kingston, P. (2002) Growing older in socially deprived areas: Social exclusion in later life, London: Help the Aged.

Understanding the risks of social exclusion in older age Appendices 60

5 Annex A: Tables 5.1 Markers of risk by age and sex Table 5-1 Low income, by age and sex

Base: All respondents ELSA

Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 87 78 74 82 Yes 13 22 26 18 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 83 69 55 73 Yes 17 31 45 27 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-2 Material deprivation, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 97 95 90 95 Yes 3 5 10 5 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 99 96 85 95 Yes 1 4 15 5 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-3 Pension wealth, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 100 100 100 100 Yes 0 0 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 96 91 81 91 Yes 4 9 19 9 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498


Table 5-4 Access to services, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 92 90 79 89 Yes 8 10 21 11 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 91 85 68 85 Yes 9 15 32 15 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-5 Transport access, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 98 96 90 96 Yes 2 4 10 4 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 98 94 79 93 Yes 2 6 21 7 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-6 Financial services, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 97 97 97 97 Yes 3 3 3 3 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 97 97 94 97 Yes 3 3 6 3 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498


Table 5-7 Social support, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 85 84 82 84 Yes 15 16 18 16 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 89 79 67 81 Yes 11 21 33 19 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-8 Contact with other people, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 85 87 92 87 Yes 15 13 8 13 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 94 94 95 94 Yes 6 6 5 6 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-9 Literacy and numeracy skills, by age and sex


Age 60-69 70-79 80+ Total



Table 5-10 Political efficacy (not voted), by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 83 90 92 87 Yes 17 10 8 13 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 85 86 85 86 Yes 15 14 15 14 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-11 Poor general health, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 91 92 92 92 Yes 9 8 8 8 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 95 91 91 93 Yes 5 9 9 7 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-12 Poor emotional health, by age and sex


Age 60-69 70-79 80+ Total



Table 5-13 Low physical activity, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 95 96 88 94 Yes 5 4 12 6 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 97 94 80 93 Yes 3 6 20 7 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-14 Bad housing, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 82 84 87 83 Yes 18 16 13 17 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 84 83 85 84 Yes 16 17 15 16 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-15 Sense of belonging in local area, by age and sex


Age 60-69 70-79 80+ Total



Table 5-16 Fear of local area after dark, by age and sex


Age 60-69 70-79 80+ Total

Marked by risk % % % % Men No 75 74 72 74 Yes 25 26 28 26 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women No 72 68 60 69 Yes 28 32 40 31 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 5.2 Socio-demographic characteristics of ELSA sample Table 5-17 Marital status, by age and sex


Age 60-69 70-79 80+ Total

Family type % % % % Men Single, never married 6 5 2 5 Married 68 70 56 67 Remarried 14 10 8 12 Legally separated or divorced 8 4 1 6 Widowed 4 11 33 10 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Single, never married 3 5 6 4 Married 60 48 17 48 Remarried 13 4 3 8 Legally separated or divorced 12 7 1 8 Widowed 12 36 72 32 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-18 Lives alone, by age and sex


Age 60-69 70-79 80+ Total

Lives alone % % % % Men Yes 12 16 27 15 No 88 84 73 85 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Yes 18 38 69 34 No 82 62 31 66 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498


Table 5-19 Number of children, by age and sex


Age 60-69 70-79 80+ Total

Children % % % % Men Yes 87 91 90 89 No 13 9 10 11 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Yes 91 88 85 89 No 9 12 15 11 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-20 Number of siblings, by age and sex


Age 60-69 70-79 80+ Total

Siblings % % % % Men Yes 79 73 61 75 No 21 27 39 25 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Yes 80 76 52 73 No 20 24 48 27 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-21 Qualifications, by age and sex


Age 60-69 70-79 80+ Total

Qualifications % % % % Men Yes 79 73 61 75 No 21 27 39 25 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Yes 80 76 52 73 No 20 24 48 27 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498


Table 5-22 Housing tenure, by age and sex


Age 60-69 70-79 80+ Total

Housing tenure % % % % Men Owner 66 79 78 72 Buyer 20 5 3 12 Private renter 11 13 16 12 Social renter or rent free 3 3 3 3 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Owner 72 72 67 71 Buyer 15 5 2 9 Private renter 10 19 25 16 Social renter or rent free 3 4 6 4 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-23 Limiting longstanding illness, by age and sex


Age 60-69 70-79 80+ Total

Limiting longstanding illness % % % % Men Yes 30 38 44 34 No 70 62 56 66 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Yes 31 41 47 38 No 69 59 53 62 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-24 Housing tenure, by age and sex


Age 60-69 70-79 80+ Total

Housing tenure % % % % Men Owner 66 79 78 72 Buyer 20 5 3 12 Private renter 11 13 16 12 Social renter or rent free 3 3 3 3 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Owner 72 72 67 71 Buyer 15 5 2 9 Private renter 10 19 25 16 Social renter or rent free 3 4 6 4 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498


Table 5-25 Index of Multiple Deprivation Quintile, by age and sex


Age 60-69 70-79 80+ Total

IMD Quintile % % % % Men Least deprived quintile 24 25 28 25 Second quintile 25 23 28 25 Third quintile 19 23 19 20 Fourth quintile 19 17 16 18 Most deprived quintile 14 11 9 12 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Least deprived quintile 25 23 21 23 Second quintile 25 24 27 25 Third quintile 21 21 18 20 Fourth quintile 18 18 20 18 Most deprived quintile 12 14 15 13 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-26 Well-being, by age and sex


Age 60-69 70-79 80+ Total

Well-being % % % % Men High 80 80 72 79 Low 20 20 28 21 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women High 84 80 70 80 Low 16 20 30 20 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-27 Carer, by age and sex


Age 60-69 70-79 80+ Total

Carer % % % % Men Yes 88 89 94 89 No 12 11 6 11 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women Yes 79 86 95 84 No 21 14 5 16 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498


Table 5-28 Population density, by age and sex


Age 60-69 70-79 80+ Total

Population density % % % % Men City 74 72 75 73 Town 12 12 12 12 Village or hamlet 15 16 13 15 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women City 75 75 76 75 Town 11 13 13 12 Village or hamlet 14 12 11 13 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 Table 5-29 Ethnicity, by age and sex


Age 60-69 70-79 80+ Total

Ethnicity % % % % Men White 98 99 99 98 Non-white 2 1 1 2 Weighted bases 1007 701 248 1956 Unweighted bases 1026 747 252 2025 Women White 98 99 100 99 Non-white 2 1 1 Weighted bases 1114 837 441 2392 Unweighted bases 1230 874 394 2498 5.3 Exploring multidimensional social exclusion Table 5-30 Multidimensional social exclusion, by sex


Sex Men Women Total

Risk marker Clusters % % % No risk markers 26 20 23 1 risk marker 30 26 28 Multi: Lonely 10 9 9 Multi: Fear of local area 11 13 12 Multi: Low income 12 20 16 Multi: Poor Health 7 7 7 Multi: Poor Access 4 6 5 Weighted bases 1956 2392 4348 Unweighted bases 2025 2498 4523


Table 5-31 Multidimensional social exclusion, by age


Age 60-69 70-79 80+ Total

Risk marker clusters % % % % No risk markers 28 21 9 23 1 risk marker 31 28 19 28 Multi: Lonely 10 9 8 9 Multi: Fear of local area 12 11 12 12 Multi: Low income 10 19 28 16 Multi: Poor Health 7 7 7 7 Multi: Poor Access 2 5 16 5 Weighted bases 2121 1538 690 4348 Unweighted bases 2256 1621 646 4523 Table 5-32 Multidimensional social exclusion, by marital status


Married Non-married Total Married Remarried Total Single Divorced/

Separated Widowed Total

Risk maker clusters % % % % % % % % No risk markers 29 25 29 2 11 16 11 23 1 risk marker 33 29 32 20 19 22 18 28 Multi: Lonely 7 13 8 22 12 13 10 9 Multi: Fear of local area 11 12 11 17 14 14 13 12 Multi: Low income 11 9 11 27 27 21 28 16 Multi: Poor Health 7 11 7 3 7 8 7 7 Multi: Poor Access 2 1 2 10 10 6 12 5 Weighted bases 2463 418 2881 197 1465 309 960 4348 Unweighted bases 2559 446 3005 196 1517 343 978 4523 Table 5-33 Multidimensional social exclusion, by whether lives alone

Base: All non-married respondents ELSA

Lives alone Yes No Total

Risk maker clusters % % % No risk markers 10 14 11 1 risk marker 18 24 19 Multi: Lonely 12 12 12 Multi: Fear of local area 14 13 14 Multi: Low income 29 20 27 Multi: Poor Health 6 10 7 Multi: Poor Access 12 7 10 Weighted bases 1105 360 1465 Unweighted bases 1156 361 1517


Table 5-34 Multidimensional social exclusion, by whether has living siblings


Has living siblings Yes No Total

Risk maker clusters % % % No risk markers 21 23 23 1 risk marker 27 28 28 Multi: Lonely 10 9 9 Multi: Fear of local area 12 12 12 Multi: Low income 17 16 16 Multi: Poor Health 6 7 7 Multi: Poor Access 7 4 5 Weighted bases 1132 3209 4348 Unweighted bases 1171 3345 4523 Table 5-35 Multidimensional social exclusion, by whether has qualifications


Has qualifications Yes No Total



Table 5-36 Multidimensional social exclusion, by tenure


Tenure Own Rent Total

Risk maker clusters % % % No risk markers 26 6 23 1 risk marker 30 16 28 Multi: Lonely 9 10 9 Multi: Fear of local area 11 15 12 Multi: Low income 14 27 16 Multi: Poor Health 6 13 7 Multi: Poor Access 4 12 5 Weighted bases 3576 766 4348 Unweighted bases 3790 728 4523 Table 5-37 Multidimensional social exclusion, by income quintile


IMD Quintile Lowest income

quintile 2nd quintile 3rd quintile 4th quintile Highest income

IMD quintile Total Risk maker clusters % % % % % % No risk markers 18 25 29 41 23 1 risk marker 18 27 30 31 33 28 Multi: Lonely 1 11 13 13 9 9 Multi: Fear of local area 17 17 14 12 12 Multi: Low income 69 12 16 Multi: Poor Health 6 9 9 8 3 7 Multi: Poor Access 5 6 6 5 2 5 Weighted bases 870 870 869 868 870 4348 Unweighted bases 861 881 899 914 968 4523 Table 5-38 Multidimensional social exclusion, by whether has limiting

longstanding illness


Has LLI Yes No Total



Table 5-39 Multidimensional social exclusion, by IMD quintile


IMD Quintile Most deprived

IMD quintile 2nd quintile 3rd quintile 4th quintile Least deprived

IMD quintile Total Risk maker clusters % % % % % % No risk markers 10 14 23 25 33 23 1 risk marker 19 25 32 30 30 28 Multi: Lonely 10 10 8 9 10 9 Multi: Fear of local area 17 15 10 10 10 12 Multi: Low income 22 21 16 16 10 16 Multi: Poor Health 14 8 6 6 4 7 Multi: Poor Access 9 7 5 4 3 5 Weighted bases 558 783 889 1075 1041 4348 Unweighted bases 546 793 921 1138 1124 4523 Table 5-40 Multidimensional social exclusion, by well-being


Well-being High Low Total

Risk maker clusters % % % No risk markers 27 8 23 1 risk marker 31 16 28 Multi: Lonely 9 12 9 Multi: Fear of local area 12 13 12 Multi: Low income 16 15 16 Multi: Poor Health 3 21 7 Multi: Poor Access 2 15 5 Weighted bases 3403 862 4348 Unweighted bases 3572 871 4523 Table 5-41 Multidimensional social exclusion, by whether is a carer

Base: All aged 60-69 ELSA

Carer Yes No Total

Risk maker clusters % % % No risk markers 28 31 28 1 risk marker 31 32 31 Multi: Lonely 10 11 10 Multi: Fear of local area 12 13 12 Multi: Low income 11 7 10 Multi: Poor Health 7 6 7 Multi: Poor Access 2 0 2 Weighted bases 1770 351 2121 Unweighted bases 1875 381 2256 A Note based on 60-69s because caring is strongly related to age.


Table 5-42 Multidimensional social exclusion, by population density


Population density City Town Village Total

Risk maker clusters % % % % No risk markers 21 25 29 23 1 risk marker 27 28 32 28 Multi: Lonely 9 9 10 9 Multi: Fear of local area 13 10 7 12 Multi: Low income 17 16 12 16 Multi: Poor Health 7 6 7 7 Multi: Poor Access 5 6 4 5 Weighted bases 3221 524 601 4348 Unweighted bases 3325 558 639 4523 5.4 Profile of the balanced panel sample Table 5-43 Age and sex profile of the balanced panel sample Base: Balanced panel sample BHPS Survey year

1997 1998 1999 2000 2001 2002 2003 2004 2005 Sex % % % % % % % % % Men 60-69 62 56 48 43 37 30 24 17 9 70-79 33 38 44 46 49 53 55 57 62 80+ 5 6 8 10 14 17 21 26 29 Women 60-69 57 51 46 41 35 29 23 17 10 70-79 38 41 44 47 48 51 52 53 54 80+ 5 8 10 13 17 20 24 30 36 Total 60-69 59 53 47 42 36 29 24 17 10 70-79 36 40 44 47 49 52 54 55 57 80+ 5 7 9 12 16 19 23 28 33 Unweighted basesB Men 412 412 412 412 412 412 412 412 412 Women 575 575 575 575 575 575 575 575 575 B Note balanced panel analysis was unweighted.


Table 5-44 Marital status, by sex and wave Base: Balanced panel sample BHPS Survey year

1997 1998 1999 2000 2001 2002 2003 2004 2005 Sex % % % % % % % % % Men Married 80 80 78 77 76 75 75 74 71 Separated or Divorced 6 6 6 5 5 5 5 5 5 Widowed 9 9 12 13 14 15 16 17 19 Never married 5 5 5 5 5 5 5 5 5 Women Not marked by risk 53 51 49 48 47 46 43 42 41 Separated or Divorced 7 7 7 7 7 7 7 7 7 Widowed 35 36 39 39 40 42 44 46 47 Never married 6 6 6 6 6 6 6 5 5 Total Not marked by risk 64 63 61 60 59 58 57 55 53 Separated or Divorced 7 7 6 6 6 6 6 6 6 Widowed 24 25 27 28 29 30 32 33 36 Never married 5 5 5 5 5 5 5 5 5 Unweighted bases Men 412 412 412 412 412 412 412 412 412 Women 575 575 575 575 575 575 575 575 575 B Note balanced panel analysis was unweighted. Table 5-45 Employment in the household, by sex and wave Base: Balanced panel sample BHPS Survey year

1997 1998 1999 2000 2001 2002 2003 2004 2005 Sex % % % % % % % % % Men No one in employment 36 34 33 28 24 21 21 17 16 One person in employment 64 66 67 72 76 79 79 83 84 Women No one in employment 26 23 22 18 16 14 14 13 11 One person in employment 74 77 78 82 84 86 86 87 89 Total No one in employment 30 28 26 22 19 17 17 14 13 One person in employment 70 72 74 78 81 83 83 86 87 Unweighted bases Men 412 412 412 412 412 412 412 412 412 Women 575 575 575 575 575 575 575 575 575 B Note balanced panel analysis was unweighted.


Table 5-46 Population density, by sex at wave 1997 Base: Balanced panel sample BHPS

Population density City Town Village or hamlet

Sex % % % Men 71 15 14 Unweighted bases 274 58 54 Women 77 11 12 Unweighted bases 403 58 63 Total 74 13 13 Unweighted bases 677 116 116 B Note balanced panel analysis was unweighted. Table 5-47 Index of Multiple Deprivation, by sex at wave 1997 Base: Balanced panel sample

IMD2000 Most deprived

IMD rank quintile

2nd quintile 3rd quintile 4th quintile Least deprived IMD rank

quintile Sex % % % % % Men 20 19 21 20 19 Unweighted bases 71 68 74 72 68 Women 25 20 22 17 15 Unweighted bases 125 99 109 85 74 Total 23 20 22 18 17 Unweighted bases 196 171 183 157 142 B Note balanced panel analysis was unweighted.


5.5 Panel sample marker of risk 1997-2005 Table 5-48 Balanced panel sample: prevalence of risk markers

Base: Balanced panel sample BHPS

Proportion marked by risk 1997 1998 1999 2000 2001 2002 2003 2004 2005 Total

Risk marker % % % % % % % % % %

Men Low contact with others 53 52 50 56 52 56 50 53 56 53 Low social support 42 40 43 40 42 38 39 33 37 39 No qualifications 39 39 39 39 39 39 39 39 39 39 Bad housing 25 26 26 24 27 25 22 22 20 24 Relative low income 18 22 21 26 25 26 27 28 29 25 Poor access to transport 18 18 19 19 21 21 22 24 25 21 Poor emotional health 12 11 9 12 11 11 15 15 16 12 Material deprivation 8 8 9 8 6 8 7 5 7 7 Poor general health 8 9 6 9 9 10 10 12 12 10 Low area satisfaction 6 5 5 6 5 5 5 5 5 5 Unweighted bases 412 412 412 412 412 412 412 412 412 3708

Women No qualifications 54 54 54 54 54 54 53 53 53 54 Low contact with others 51 50 53 53 47 54 50 55 54 52 Poor access to transport 38 39 40 41 43 43 43 46 48 42 Low social support 36 33 36 35 40 34 34 28 34 34 Relative low income 31 32 29 34 35 32 32 32 33 32 Bad housing 26 25 26 25 24 24 21 21 20 24 Poor emotional health 15 17 18 19 20 20 21 23 25 20 Material deprivation 9 8 7 7 5 6 5 5 6 7 Poor general health 9 9 5 10 12 13 14 13 12 11 Low area satisfaction 6 5 6 6 6 6 5 5 5 6 Unweighted bases 575 575 575 575 575 575 575 575 575 5175

Total Low contact with others 51 50 52 54 49 55 50 54 55 52 No qualifications 48 48 48 48 48 48 48 48 47 48 Low social support 38 36 39 37 41 36 36 30 35 36 Poor access to transport 30 30 31 32 34 34 35 37 38 33 Bad housing 26 26 26 24 26 24 21 21 20 24 Relative low income 26 28 26 31 31 30 30 30 32 29 Poor emotional health 13 15 15 16 16 16 18 19 21 17 Material deprivation 9 8 8 7 6 7 6 5 7 7 Poor general health 9 9 5 10 11 12 12 13 12 10 Low area satisfaction 6 5 5 6 6 5 5 5 5 5 Total unweighted bases 987 987 987 987 987 987 987 987 987 8883 B Note balanced panel analysis was unweighted.


5.6 Risk marker prevalence within clusters

Table 5-49 Dynamic social exclusion clusters, by year

Base: Balanced panel sample experiencing multiple risk markers BHPS

Multi: Low support & loneliness

Multi: Low relative income and transport

Multi: Health problems and loneliness Longitudinal social exclusion

Clusters % % % 1997 Low relative income 6 70 20 Material deprivation 2 24 10 Poor access to transport 17 75 14 Low social support 70 27 50 Low contact with others 73 44 68 No qualifications 60 68 44 Poor general health 8 43 Poor emotional health 7 77 Bad housing 37 33 30 Unweighted bases 286 287 139 1998 Low relative income 9 74 29 Material deprivation 2 24 6 Poor access to transport 35 69 22 Low social support 59 46 44 Low contact with others 62 39 55 No qualifications 73 76 55 Poor general health 9 45 Poor emotional health 6 83 Bad housing 66 23 32 Unweighted bases 174 256 139 1999 Low relative income 10 70 19 Material deprivation 1 24 7 Poor access to transport 25 73 26 Low social support 80 30 53 Low contact with others 52 49 54 No qualifications 70 75 46 Poor general health 10 4 16 Poor emotional health 9 91 Bad housing 50 33 31 Low area satisfaction 12 5 13 Unweighted bases 212 267 112 2000 Low relative income 9 74 37 Material deprivation 2 22 5 Poor access to transport 36 70 25 Low social support 58 44 46 Low contact with others 69 44 57 No qualifications 68 78 48 Poor general health 7 50 Poor emotional health 7 82 Bad housing 56 26 29 Low area satisfaction 13 6 13 Unweighted bases 166 266 157 2001 Low relative income 14 73 28 Material deprivation 1 16 7 Poor access to transport 34 73 33 Low social support 88 33 62 Low contact with others 58 44 56 No qualifications 63 75 52 Poor general health 8 53 Poor emotional health 1 8 82 Bad housing 40 35 39 Low area satisfaction 11 7 12 Unweighted bases 173 269 153


Table 5-49 continued Dynamic social exclusion clusters, by year Base: Balanced panel sample experiencing multiple risk markers BHPS

Multi: Low support & loneliness

Multi: Low relative income and transport

Multi: Health problems and loneliness Longitudinal social exclusion

Clusters % % % 2002 Low relative income 11 73 27 Material deprivation 1 21 5 Poor access to transport 43 69 31 Low social support 54 44 48 Low contact with others 71 48 57 No qualifications 70 75 57 Poor general health 8 54 Poor emotional health 7 75 Bad housing 55 22 37 Low area satisfaction 8 6 13 Unweighted bases 164 260 166 2003 Low relative income 13 72 26 Material deprivation 2 16 6 Poor access to transport 40 73 36 Low social support 86 25 51 Low contact with others 57 45 57 No qualifications 70 78 51 Poor general health 11 52 Poor emotional health 10 83 Bad housing 37 23 32 Low area satisfaction 9 6 9 Unweighted bases 149 264 169 2004 Low relative income 9 72 30 Material deprivation 2 15 3 Poor access to transport 45 74 36 Low social support 51 32 44 Low contact with others 59 42 60 No qualifications 72 77 48 Poor general health 9 57 Poor emotional health 1 9 80 Bad housing 55 18 28 Low area satisfaction 13 3 10 Unweighted bases 158 265 176 2005 Low relative income 13 70 30 Material deprivation 1 17 6 Poor access to transport 43 75 37 Low social support 83 25 47 Low contact with others 58 47 63 No qualifications 66 76 47 Poor general health 9 49 Poor emotional health 3 13 84 Bad housing 34 22 28 Low area satisfaction 11 5 10 Unweighted bases 157 283 188 Total Low relative income 13 70 30 Material deprivation 1 17 6 Poor access to transport 43 75 37 Low social support 83 25 47 Low contact with others 58 47 63 No qualifications 66 76 47 Poor general health 9 49 Poor emotional health 3 13 84 Bad housing 34 22 28 Low area satisfaction 11 5 10 Unweighted bases 1639 2417 1399


5.7 Socio-demographic characteristics of dynamic social exclusion clusters

B Note balanced panel analysis was unweighted. Table 5-51 Dynamic social exclusion clusters, by age Base: Balanced panel sample BHPS

Age group 60-69 70-79 80-89 Total Longitudinal social exclusion

Clusters % % % % No risk markers 18 11 7 13 1 risk marker 29 26 19 26 Multi: social support & lonely 21 17 17 18 Multi: low income & transport 18 30 40 27 Multi: health & low social support 16 15 17 16 Unweighted bases 3122 4270 1491 8883 B Note balanced panel analysis was unweighted. Table 5.52 Dynamic social exclusion clusters, by family type Base: Balanced panel sample BHPS

Family type Living alone Living as a couple Living with others Total Longitudinal social exclusion

Clusters % % % % No risk markers 8 17 5 13 1 risk marker 19 30 20 26 Multi: social support & lonely 16 19 30 18 Multi: low income & transport 42 19 20 27 Multi: health & low social support 15 15 25 16 Unweighted bases 3099 5354 430 8883 B Note balanced panel analysis was unweighted. Table 5-53 Dynamic social exclusion clusters, by marital status Base: Balanced panel sample BHPS

Marital status Married Divorced or

separated Widowed Never married Total

Longitudinal social exclusion Clusters % % % % % No risk markers 16 6 9 7 13 1 risk marker 30 23 20 21 26 Multi: social support & lonely 19 19 17 19 18 Multi: low income & transport 20 37 39 34 27 Multi: health & low social support 15 17 16 19 16 Unweighted bases 5228 569 2620 466 8883 B Note balanced panel analysis was unweighted.

Table 5-50 Dynamic social exclusion clusters, by sex


Sex Men Women Total Longitudinal social exclusion

Clusters % % % No risk markers 17 10 13 1 risk marker 29 23 26 Multi: social support & lonely 20 17 18 Multi: low income & transport 21 31 27 Multi: health & low social support 13 18 16 Unweighted bases 3708 5175 8883


Table 5-54 Dynamic social exclusion clusters, by housing tenure Base: Balanced panel sample BHPS

Tenure Own home Buying with the

help of a mortgage

Social renter Total

Longitudinal social exclusion Clusters % % % % No risk markers 16 3 8 13 1 risk marker 30 11 19 26 Multi: social support & lonely 17 22 23 18 Multi: low income & transport 22 47 28 27 Multi: health & low social support 15 17 23 16 Unweighted bases 6804 1894 185 8883 B Note balanced panel analysis was unweighted. Table 5-55 Dynamic social exclusion clusters, by population density at first wave Base: Balanced panel sample at first wave (1997) BHPS

Population density City Town Village or Hamlet Total Longitudinal social exclusion

Clusters % % % % No risk markers 7 10 11 13 1 risk marker 18 22 27 26 Multi: social support & lonely 29 29 34 18 Multi: low income & transport 32 20 15 27 Multi: health & low social support 14 18 13 16 Unweighted bases 677 116 116 909 B Note balanced panel analysis was unweighted. Table 5.56 Dynamic social exclusion clusters, by Index of Multiple Deprivation at first wave Base: Balanced panel sample at first wave (1997) BHPS

IMD2000 Most deprived

IMD rank quintile

2nd quintile 3rd quintile 4th quintile Least deprived IMD rank

quintile

Total

Longitudinal social exclusion Clusters % % % % % % No risk markers 7 6 8 6 13 13 1 risk marker 10 18 22 27 27 26 Multi: social support & lonely 32 35 20 36 31 18 Multi: low income & transport 42 27 33 17 17 27 Multi: health & low social support 10 14 17 15 12 16 Unweighted bases 196 171 183 157 142 849 B Note balanced panel analysis was unweighted.


5.8 Duration of multidimensional social exclusion Table 5-57 Duration of low social support and loneliness

multiple risk markers Base: All in low social support and loneliness cluster on at least one occasion

Multidimensional cluster

Low social support and loneliness

Occasions in cluster % 1 13 2 16 3 11 4 16 5 16 6 10 7 10 8 8 9 - Unweighted bases 822

Table 5-58 Duration of low income and transport multiple risk

markers Base: All in low income and transport cluster on at least one occasion


Low income and transport

Occasions in cluster % 1 18 2 14 3 11 4 11 5 8 6 10 7 10 8 7 9 10 Unweighted bases 631

Table 5-59 Duration of health and loneliness multiple risk

markers Base: All in low income and transport cluster on at least one occasion


Health and loneliness

Occasions in cluster % 1 26 2 22 3 18 4 14 5 8 6 6 7 4 8 3 9 - Unweighted bases 642


5.9 Persistence in clusters: cluster transitions Low social support and loneliness cluster

Table 5-60 Persistence and movement of Low social support and loneliness


Average wave by wave transition t-1 t t+1

Longitudinal Risk Marker Clusters % % % No risk markers 1 1 1 risk marker 16 15 Multi: social support & lonely 45 100 44 Multi: low income & transport 25 26 Multi: health & low social support 12 13

Table 5-61 Persistence and movement of Low income and transport Problems



Longitudinal Risk Marker Clusters % % % No risk markers 0 0 1 risk marker 8 6 Multi: social support & lonely 17 16 Multi: low income & transport 67 100 68 Multi: health & low social support 8 9

Table 5-62 Persistence and movement of Health and loneliness



Longitudinal Risk Marker Clusters % % % No risk markers 3 2 1 risk marker 16 16 Multi: social support & lonely 14 14 Multi: low income & transport 16 14 Multi: health & low social support 51 100 54


6 Annex B: Technical 6.1 Correlation analysis Correlation analysis and latent class analysis, was used to explore the strong and weak associations between the 16 markers of risk experienced by older people. Tetrachoric correlations between the 16 risk markers were investigated, and the resulting correlation matrix can be seen in Table 5-1. Correlations higher than 0.2 have been highlighted in table 1.1. (Tetrachoric correlations assume a latent bivariate normal distribution for each pair of variables). Findings indicate that some of the 16 ELSA markers of risk correlate particularly highly with each other, while others do not. The highest correlation is between poor general health and not being physically active at 0.61. The results of the tetrachoric correlation analysis show the potential for some interesting overlapping risk markers, as some of the 16 risk marker from different themes in the B-SEM correlate together, for example low social support and material deprivation correlate together at 0.41, and poor general health and poor access to services are correlated at 0.45. The correlation matrix suggests that risk markers experienced by older people across the 16 Risk marker are multidimensional, and the experience is different for different people.


Table 6-1 Tetrachoric Correlation Matrix

Rel

ativ

e lo

w in

com

e am

ong

thos

e ag

ed 6

0 pl

us

Mat

eria

l dep

rivat

ion

No

pens

ion

wea

lth

Poo

r acc

ess

to

serv

ices

P

oor a

cces

s to

tra

nspo

rt

No

finan

cial

ser

vice

s

Low

soc

ial s

uppo

rt

Non

e or

Low

con

tact

w

ith o

ther

s

Poo

r lite

racy

and

nu

mer

acy

skill

s

Low

pol

itica

l pa

rtici

patio

n

Poo

r gen

eral

hea

lth

Poo

r em

otio

nal h

ealth

Not

phy

sica

lly a

ctiv

e

Hou

sing

pro

blem

s

Sen

se o

f not

be

long

ing

Fear

of w

alki

ng a

lone

af

ter d

ark

Relative low income among those aged 60 plus 1.00 Material deprivation 0.22 1.00 No pension wealth 0.41 1.00

Poor access to services 0.08 0.22 0.12 1.00

Poor access to transport 0.20 0.35 0.28 0.55 1.00 No financial services 0.27 0.21 0.32 0.16 0.16 1.00 Low social support 0.17 0.41 0.16 0.20 0.35 1.00

Low contact with others 0.12 -0.15 0.17 1.00 Poor literacy and numeracy skills 0.15 0.24 0.12 0.17 0.22 0.27 0.10 1.00

Low political participation 0.12 0.13 0.10 0.10 0.12 0.09 1.00 Poor general health 0.45 0.40 0.28 0.15 0.22 0.15 1.00 Poor emotional health 0.12 0.12 0.39 0.29 0.25 0.19 0.19 0.14 0.55 1.00 Not physically active 0.22 0.17 0.51 0.54 0.34 0.15 0.10 0.18 0.13 0.61 0.35 1.00 Housing problems 0.07 0.12 0.12 0.15 0.17 0.15 0.12 0.08 0.20 0.23 0.09 1.00

Sense of not belonging 0.13 0.12 0.17 0.19 0.18 1.00 Fear of walking alone after dark 0.11 0.10 0.16 0.19 0.13 0.13 0.18 0.15 0.15 0.15 0.16 1.00


6.2 Detailing the Latent Class Analysis 6.2.1 Latent Class Analysis LCA is a statistical technique that can be used to identify relationships in survey data when respondents’ answers to questions are categorical. An LCA analysis divides respondents into groups (or latent classes) on the basis of their answers to a series of questions. The aim is for each class to be reasonably homogeneous, in that every individual in a class is assumed to be similar (in the sense of having the same response probabilities for each question) while respondents in different classes are assumed to be dissimilar. Applied to the ELSA markers of risk, LCA allows us to investigate whether there are discrete groups of individuals who experience similar forms of multiple risk markers. Once groups such as these are found, the analysis generates a probability for each respondent of them being in each class and assigns them to the class for which they have the highest probability of membership. It will also usually be possible to relate membership of each class with the respondent’s answers to each question and thus describe each class. This is not a straightforward task, but it can be done either by using the output from the LCA program or by performing a further analysis on the data in another package. In Section 2.2 we discuss five clusters identify by the LCA analysis. In the next section of this technical report we will describe how this number of classes was identified. 6.2.2 Latent GOLD The data were modelled using the package Latent GOLD26, a software package that can implement several types of latent class models. A useful feature of Latent GOLD is that it is compatible with packages such as SPSS. In the analysis of multiple risk markers we read the data from SPSS, used Latent GOLD to identify the classes and then exported the results back in to SPSS for further analyses. As a result, we were able to create an SPSS file with variables for: • the respondent’s serial number; • the markers of risk experienced by respondents; • the probability assigned to each individual of them being in each class; and • the class for which they have the highest probability of membership. A typical analysis involved fitting several models with different numbers of classes. It was then possible to write SPSS syntax to compare different models – for example to compare a model containing five classes with one containing six. This allowed us to identify the most useful model.

26 See the user’s guide for a full description: Vermunt, J.K. and Magidson, J. (2005) Latent GOLD 4.0 User’s Guide. Belmont, Massachusetts: Statistical Innovations Inc.


6.2.3 Modelling multiple markers of risk

Features of the data LCA can be used to model any data set where response variables are categorical (either nominal or ordinal) which has an underlying nominal latent variable to define latent classes. As the 16 markers of risk are nominal, it is amenable to analysis by LCA. Nevertheless, there are certain features of the data that required particular attention before using LCA: • The ELSA data set is sparse. The data consist of 16 risk markers, each with 2 possible

answers. This gives 216=65,536 possible patterns of answers. There are only 2,166 respondents experiencing multiple risk markers, so only a small proportion of the possible response patterns can be attained. A consequence of this is that many of the standard test statistics produced by LCA packages to compare a k-cluster model with a (k+1)-cluster model will not be valid and the user should treat the results of these tests with caution27. Although the sample size is such that standard statistical test can not be used to choose between models, once a model is chosen, provided the number of clusters is not too large, the sample size is sufficiently large to allow a good description of each cluster.

• The large number of questions used can also cause computational problems if a large number of clusters are fitted. With 16 risk markers, each with two possible answers, a k-cluster model involves estimating 1*16*k+(k-1) parameters. It is possible that the programme will fail to find a solution (or find an incorrect solution) if k is large. Advice on how to guard against this is given in the Latent GOLD technical guide.

Identifying the number of classes As part of a Latent Class Analysis we need to identify the number of classes. In practice, it is unlikely that there will be a single "correct" model so it is usual to consider a range of possible models containing different numbers of classes and choose the most appropriate using some criteria. A general approach to statistical model fitting is to try to balance the fit and the parsimony of a model – generally if two models fit a data set equally well the one with fewer parameters will be chosen. Under this principle, in LCA, if a model with k+1 classes fits the data just as well as one with k classes the k-class model will be chosen. LCA software packages such as Latent GOLD provide the analyst with statistics to help in the choice of the correct number of classes in the data. In particular a process analogous to a forward selection procedure in regression modelling is sometimes used. The process starts by fitting a one-class model and then adds a class at a time. A formal hypothesis test can be performed to see if a k+1-class model is an improvement on a k-class model. (The null hypothesis is that the k-class model generates homogeneous classes; the alternative hypothesis is that the k+1-class model

27 Latent GOLD calculates a statistic, L2, which is similar to a chi-squared statistic but the help system warns: “with sparse data, the chi-squared based estimation for the p-value associated with L2 cannot be trusted because these statistics do not follow a chi-squared distribution.” The reason for this is that chi-squared tests are only valid if expected cell sizes are not small (greater than 5 is a common cut-off point). With a data set as sparse as this many of the expected cell sizes will be far too small to allow use of either the standard chi-squared statistic or L2.


gives significantly more homogeneity.) If the test is statistically significant the k+1-class model is considered as being the preferred model. The process continues until adding a class does not lead to a statistically significant improvement. This procedure can be performed wholly within Latent GOLD. However, there are two objections to this approach when applied to the multiple risk markers data. A technical problem is that mentioned above: the p-values calculated by the package are not valid when analysing a data set as sparse as the multiple risk markers data28. A second problem is that the size of the data set (16 questions) is large enough to mean that the significance tests might not be very powerful. Even when classes display a large difference on one or two questions the overall significance test will be found to be “not significant” if the classes are similar on the other questions. In other words, with a very large number of risk markers it is almost impossible to generate a small number of classes with homogeneity within these classes. As a result of this, the standard test statistics given by the package are of limited use and other means of testing that the classes are a reasonable summary of the data are needed. This means that if an automatic selection routine is to be used (as we have done), then, rather than choosing a model on the basis of the p-values obtained from a formal hypothesis test, we recommend using an informal assessment. Part of this assessment can be based on a goodness-of-fit measure. A goodness-of-fit measure can be calculated for both the k-class model and the k+1-class model, and the k+1-class model would be chosen if its goodness-of-fit measure is better than that of the k-class model. Latent GOLD provides several goodness-of-fit statistic statistics to help decide on an appropriate model. Three of these, the Bayesian Information Criterion (BIC), the Akaike Information Criterion (AIC) and the Akaike Information Criterion 3 (AIC3), are shown in the table below for multiple risk markers data.

28 The Latent GOLD User’s Guide suggests dealing with sparse data by calculating a bootstrap p-value. However, on a data set of this size this seems to be computationally intensive. It also does not overcome the second problem.


Table 6-2 Latent Class Analysis models and goodness-of-fit statistics Number of classes

BIC(LL)

AIC(LL)

AIC3(LL)

1 31832 31644 316772 31782 31498 315483 31807 31426 314934 31849 31372 314565 31882 31309 314106 31961 31291 314097 31830 31353 314378 31830 31353 31437 The interpretation of BIC, AIC and AIC3 is that small values correspond to a good fit. On a strict interpretation of these statistics the “optimal” model suggested by BIC is a 2-clusters model. AIC and AIC3 both suggest an 6-cluster model (but both suggest it’s not much of an improvement on a 5-cluster model). Thus, an automatic selection method would choose two clusters if BIC was used as the selection criterion, and five or six clusters if AIC or AIC3 were used. To decide between these options other criteria are needed: the approach we prefer is as follows: First, Latent GOLD is used to fit models with varying numbers of classes. (For example, we might start by looking at every model from 1 to 8 classes). Goodness-of-fit statistics are then examined for each of the models. Examining these statistics should allow us to rule out certain models as having too poor a fit to be considered, and also give an upper limit for the number of classes that need to be considered. On this basis the models containing five and six classes should be examined in further detail – though to be safe both the 4-class and 7-class models could be considered. The choice between these should then be made on the basis of several less formal considerations: 1. The class sizes should be examined. A statistically significant result need not be important in a

practical sense. A model with a large number of classes might result with some very small classes.

2. The membership probabilities can also be examined. Ideally each individual would have a fitted probability of 1 of being in one class and probabilities of 0 of being in the others (thus indicating that we can assign each individual to its class with complete certainty). In practice the best that can be hoped for is that these probabilities will be close to either 1 or 0. Consequently an examination of these probabilities will aid in the choice of model.

3. Where two or more models seem equally good the principle of parsimony suggests the model with fewer classes should be chosen.

4. Finally, routines can be written (for example in SPSS) to establish which of these models leads to a sensible definition of classes.

When analysing the multiple risk markers data, we found there was not much difference between the six-, and five-cluster models. The six-cluster model gave sample sizes in one class that were too small for analysis so a five-cluster model was chosen as our working model.


Classifying individuals and describing classes Once a working model has been chosen the analyst will usually try to relate membership of each class with the respondents’ answers to each question and thus describe (label) each class. One method of doing this is to examine the parameter estimates obtained by the model. Latent GOLD estimates the probability of membership of each class and the probability associated with each class for its answers to each question. For example, cluster one has a 99% probability of respondents being marked by income risk, whereas cluster 2 has 0%. Thus, cluster one will be more associated with relative low income than cluster two. Section 2.2 uses this method to define the clusters produced by the five-cluster model. Another method is to examine the responses rather than the parameters. This is the type of analysis described in detail in chapter 2.3. Either of these methods can be used to help describe classes. The first method has the advantage that it does not need to assign individuals to classes (the second method assigns respondents to their modal class and hence does not take into account the uncertainty concerning class membership). On the other hand, the second method might be preferable as its class labels are based on descriptions of a real sample rather than estimates of parameters (many of which could have quite large standard errors).


Table 6-3 Conditional probability parameters of Latent Class Analysis model

Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster Size Base 429 527 701 304 205

% 20 24 32 14 9%

Latent Class Analysis Probabilities Cluster probability 0.2086 0.2264 0.2819 0.161 0.122

Probability of cluster

membership


membership


membership


membership


membership

Relative Low income among those aged 60 plus 0.12 0.0199 0.9972 0.2407 0.2275

Material deprivation 0.0742 0.0635 0.0903 0.0013 0.1958

Low pension wealth 0.0593 0.0352 0.1381 0.0427 0.1047

Difficult access to services using usual transport 0.1766 0.1391 0.123 0.3318 0.655

No access to car or regular use of public transport 0.0315 0.044 0.0574 0.0004 0.5789

No financial services 0.0284 0.0347 0.0624 0.0915 0.0761

Low social support 0.4167 0.268 0.2762 0.1501 0.4098

Live alone and/or v. little contact with family/friends 0.3255 0.1451 0.0916 0.069 0.1202

Poor literacy and numeracy skills 0.2594 0.2952 0.2518 0.2713 0.3279

Low political engagement (not voted) 0.3395 0.201 0.1744 0.2172 0.2072

Poor self-reported health 0.0173 0.0486 0.0062 0.5026 0.3604

Poor emotional health (CESD) 0.1153 0.1322 0.1044 0.5304 0.323

Rarely or never do physical activity 0.0285 0.0422 0.0262 0.2505 0.4316

One or more housing problems 0.3997 0.2348 0.2038 0.333 0.1842

Feeling of not belonging in local area 0.2006 0.1711 0.0741 0.1195 0.0623

Fear of walking alone at night 0.0083 0.9974 0.4211 0.3199 0.4099


6.2.4 Using Latent Class Analysis longitudinally Latent Class Analysis was run on a cross-sectional sample of older people aged 60 and over, in each wave for which there was complete data (waves 1997, 2001, 2003 and 2005). The model structure produced by the analysis was similar across the four waves (Table 6-4). A three cluster model was the best fit according to the tests detailed in section 5.2.3, for each of the four waves. As the analysis produced similar results in four waves, one of these waves was used as a ‘template’ of how risk markers group together. This ‘template’ wave, from which cluster formation in subsequent waves was based, was the first wave in the analysis (1997).

Using the recruitment probabilities produced by the Latent Class Analysis for the clusters in survey year 1997, the a posteriori probability of an individual’s membership in each class was calculated for all nine waves in the analysis. In each wave of the analysis, an individual was assigned to the latent class with the highest a posteriori probability (modal assignment).

Table 6-4 Latent Class Analysis model probabilities across 4 waves of BHPS

Wave Wave Wave 1997 2001 2003 2O05 1997 2001 2003 2O05 1997 2001 2003 2O05

Low relative income

& transport Low social support &

loneliness Health problems &

loneliness Low level of social support 0.39 0.54 0.46 0.44 0.81 0.90 0.46 0.57 0.57 0.61 0.47 0.43

Hhold income < 60%

BHPS median 0.62 0.51 0.49 0.43 0.11 0.04 0.49 0.52 0.21 0.32 0.39 0.39

Material deprivation 0.24 0.12 0.22 0.21 0.03 0.00 0.22 0.04 0.09 0.07 0.06 0.06

No private transport 0.72 0.59 0.87 1.00 0.24 0.05 0.87 0.13 0.18 0.38 0.53 0.51

Housing problems 0.36 0.32 0.20 0.18 0.53 0.97 0.20 0.37 0.33 0.42 0.31 0.27

Low contact with people 0.43 0.54 0.54 0.53 0.58 0.63 0.54 0.53 0.63 0.42 0.55 0.60

No qualifications 0.74 0.74 0.72 0.67 0.73 0.18 0.72 0.76 0.47 0.45 0.63 0.52

Poor general health 0.11 0.12 0.00 0.07 0.00 0.05 0.00 0.14 0.39 0.52 0.48 0.38

Poor mental health 0.10 0.10 0.04 0.04 0.01 0.15 0.04 0.03 0.68 0.98 0.57 0.92

Low neighbourhood satisfaction 0.06 0.07 0.10 0.05 0.15 0.19 0.04 0.10 0.12 0.07 0.13 0.10

Social Exclusion Task Force Cabinet Office Admiralty Arch The Mall London SW1A 2WH Telephone: 020 7276 1234 E-mail: [email protected] Web address: www.cabinet-office.gov.uk Publication date: July 2009 © Crown copyright 2009 The text in this document may be reproduced free of charge in any format or media without requiring specific permission. This is subject to the material not being used in a derogatory manner or in a misleading context. The source of the material must be acknowledged as Crown copyright and the title of the document must be included when reproduced as part of another publication or service.

Understanding the risks of social exclusion across …...The analysis and views expressed in this...

Documents

Transcript of Understanding the risks of social exclusion across …...The analysis and views expressed in this...